下载文档原格式
中华行为医学与脑科学杂志2020年12月第29卷第12期Chin J Behav Med Brain Sci,December2020,Vol.29,No.12・1143•・系统综述・精神分裂症执行功能与脑影像学魏倩倩张荣荣彭小慧闫伟谢世平南京医科大学附属脑科医院精神科210029通信作者：谢世平，Email：xieshiping@.on【摘要】精神分裂症是一类病因未明确的精神障碍，认知功能障碍是精神分裂症核心症状之一,精神分裂症的预后和社会功能恢复主要取决于认知受损的程度，特别是执行功能。
脑影像学客观地反映了精神分裂症患者的脑结构和功能及代谢改变，既往神经影像学研究发现脑结构异常可能是执行功能障碍的发病基础之一，通过学习探究脑影像学,可以进一步探索精神分裂症的生物学标志物,并应用到精神分裂症的预防、诊断、治疗等方面。
本文通过结构磁共振、功能磁共振、磁共振波谱(magnetic resonance spectroscopy,MRS)、正电子发射计算机断层显像(positron emission tomography,PET)等方面来探究精神分裂症执行功能与脑影像的关系，发现精神分裂症患者在脑影像学上较健康人群发生改变，主要表现有前额叶形态异常,海马体积减小，脑白质完整性受损，前额叶等皮层激活减少，脑网络传递异常,大脑谷氨酰胺、多巴胺代谢紊乱等，且执行功能受损与其改变存在相关性。
通过研究这些精神分裂症执行功能脑影像学的改变，未来可以探究出新的精神分裂症生物学标志物，精准制定个体化治疗方案，改善患者认知障碍，促进社会功能的恢复。
【关键词】精神分裂症；执行功能；磁共振成像基金项目：国家重点研发资助项目(2016YFC1306800)；江苏省南京市卫生计生委重点项目(ZKX15033)DOI：10.3760/371468-20200614-01472Executive function and brain imaging in schizophreniaWei Qianqian.Zhang Rongrong,Peng Xiaohui,Yan Wei,Xie ShipingDepartment of Psychiatry,Affiliated Nanjing Brain Hospital,Nanjing Medical University,Nanjing210029,ChinaCorresponding author:Xie Shiping,Email：xieshiping@[Abstract]Schizophrenia is a kind of mental disorder with unclear etiology,and cognitive impairmentis one of the core symptoms of schizophrenia.The prognosis and social function recovery of schizophreniamainly depend on the degree of cognitive impairment,especially executive function.Brain imaging can objectively reflect the changes of brain structure and function and metabolic of patients with schizophrenia.Previous neuroimaging studies have found brain structure abnormalities may be one of the pathogenesis of the executive dysfunction.By leaniing brain imaging,further researches were done to explore the biological markersof schizophrenia,and were applied to the prevention,diagnosis,treatment of schizophrenia,and so on.Thepurpose of this article is to explore the relationship between executive function and brain imaging of schizophrenia by structural magnetic resonance,functional magnetic resonance spectroscopy(MRS)and positron e-mission tomography(PET).This review found that patients with schizophrenia compared with healthy peopleon brain imaging changes,mainly with abnormal morphology of prefrontal lobe,decrease of hippocampal volume,impairment of white matter integrity,decrease of activation of prefrontal cortex,abnormal transmission ofbrain network,disorder of brain glutamine and dopamine metabolism,and executive function impairmenl wascorrelated with the changes.Through investigating these changes in brain imaging of executive function ofschizophrenia,new biological markers of schizophrenia can be explored in the future,contributing to accurately formulate personalized treatment plans,improve cognitive impairment of patients and promote the recoveryof social functions.[Key words]Schizophrenia；Executive function；Magnetic resonance imagingFund programs:National Key Research and Development Funding Projects(2016YFC1306800)；KeyProject of Nanjing Health and Family Planning Commission of Jiangsu Province(ZKX15033)DOI：10.3760/371468-20200614-01472•1144-中华行为医学与脑科学杂志2020年12月第29卷第12期Chin J Behav Med Brain Sci,December2020,Vol.29,No.12精神分裂症是一种严重的、慢性致残性精神疾病，在一般人群中，终身患病率约为1%：'：。
第42 卷第 10 期2023 年10 月Vol.42 No.101388~1394分析测试学报FENXI CESHI XUEBAO（Journal of Instrumental Analysis）高分辨质谱与深度学习视阈下的藏红花品质分析宋含1，房芳2，魏哲文2，杨丽3，巩志国2*，张荣1，贾玮1*（1．陕西科技大学食品科学与工程学院，陕西西安710021；2．乌鲁木齐海关技术中心，新疆乌鲁木齐830063；3．阿拉山口海关技术中心，新疆阿拉山口833418）摘要：高分辨质谱结合传统数据分析方法常用于藏红花品质鉴别。
但传统数据分析算法的准确率低、模型拟合效果差，导致判别结果的假阳性率与假阴性率高，难以进行精准鉴别。
深度学习算法可自动提取数据中的隐藏特征信息，与高分辨质谱技术结合，可实现藏红花品质的精准鉴别。
该文简要概述了深度学习算法在处理高分辨质谱代谢组学数据方面的研究进展，并在此基础上，总结了深度学习算法在藏红花品质鉴别中的应用，探讨了高分辨质谱技术结合深度学习算法鉴别藏红花品质的可行性。
关键词：深度学习；藏红花；高分辨质谱；特征提取；品质鉴别中图分类号：O657.63；TB9文献标识码：A 文章编号：1004-4957（2023）10-1388-07Saffron Quality Analysis Based on High Resolution Mass Spectrom⁃etry and Deep Learning Visual ThresholdSONG Han1，FANG Fang2，WEI Zhe-wen2，YANG Li3，GONG Zhi-guo2*，ZHANG Rong1，JIA Wei1*（1．School of Food Science and Engineering，Shaanxi University of Science and Technology，Xi'an 710021，China；2．Urumqi Customs Technical Center，Urumqi 830063，China；3．Alashankou Customs TechnicalCenter，Alashankou 833418，China）Abstract：High resolution mass spectrometry combined with traditional data analysis methods is of⁃ten used to identify the quality of saffron. However，the traditional data analysis algorithm sometimes has low accuracy and poor model fitting effect，resulting in high false positive rate and false negative rate of discrimination results，and it is difficult to accurately identify the quality of saffron.Deep learning algorithm can automatically extract the hidden feature information in the data，and combine with high resolution mass spectrometry technology to accurately identify the quality of saffron.This paper briefly summarizes research progresses of deep learning algorithms in processing high resolution mass spectrometry metabolomics data. On this basis，applications of deep learning algorithm in saf⁃fron quality identification were summarized，and the feasibility of combining high resolution mass spectrometry with deep learning algorithm was further discussed.Key words：deep learning；saffron；high resolution mass spectrometry；feature extraction；quali⁃ty identification藏红花为高附加值产品，存在劣质原料掺假现象，而溯源分析可保障藏红花产业链高质量发展。
第44卷第4期2021年4月Vol.44Ao.4Apr.2021计算机学报CHINESE JOURNAL OF COMPUTERS基于绝句生成的构造式信息隐藏算法秦川王萌司广文姚恒（上海理工大学光电信息与计算机工程学院上海200093）摘要目前利用文本生成进行秘密信息隐藏的研究中，生成文本的质量和嵌入率是主要存在的挑战•为此本文提出一种基于绝句生成的构造式信息隐藏算法•首先对绝句文本数据进行词向量的构建，然后利用机器翻译模型搭建新的绝句诗生成模型；在诗句生成阶段，需要向模型输入主题词、模板信息和押韵信息，通过输入信息的综合作用，生成第一行绝句诗；再利用现有可体现诗句主题的显著信息生成后续行的诗句首利用绝句诗生成模型进行信息隐藏的过程中，对于相同的输入信息，模型可生成多首同一主题且符合模板要求的绝句诗；通过对主题词、模板信息、押韵信息以及诗句的不同选择,可有效实现秘密信息的隐藏首验中使用了25000首绝句诗进行训练，结果表明本文算法的嵌入率可达到35%左右，与已报道的主流算法相比具有一定的优势，且本文所提出的模型生成的诗词语义通顺、主题明确为关键词文本生成；构造式信息隐藏;绝句；主题词；嵌入率中图法分类号TP303DOI号年.为年7/SP首首16.2021.00775Constructive Information Hiding with Chinese Quatrain GenerationQIA Chuan WANG Meng PI Guang-Wen YAL Heng(School of OpLical-ElecLrical and Computer Engineering,UniversiLy of Shanghai for Science and Technology,Shanghai200093)Abstract As one kind ol the main media ol information communication,the text is one ol the common cover dato in the field ol informatioc hiding,whicO hao attracted extensivo interests inthe related researcO community.However,compared with digital image,viden anO other typeo of cover data,the redundancp of text is relativelp lowee anO the spacr foe datt embeddin-is relativelp limiteO.At present,the technique of text generatioo based on natural langua-r processin-becomee morr and morr popular,buh its applicatiog in the field of informatioo hiding n noh satisfactorp, which is still in the initial sta-e.A the curregt research of mformatioh hiding using text generatiog,the semantio qualitp of geneeed text nd the embedding ratr arr the two challenges. Therefore,in this paper,we propose n novel constructive informatioh hiding method Used on Chinesr qulain generatioh.First,word vectore arr constructed foe the41X3^00text hata,andnt the same time,the informatioh of keyword,pattere and rhyme foe the qulain poetrp need toUr determined and encoded,which then participatr in the training of the quatraid generatiod model. Then,the quatrain generation model proposed in this papee is established through the Sequencrto Sequence model,which is baset on g bidirectional recurredS neurm netword with the attentiod mechanism.A the stage of quatraid generatiod,the first line of quatraid cad Ur generated basedon the inpul keywords and the combined eSecl of pattere informatiod and rhyme informatiod. Then,Up selectino the worts related to the keyworts oe the guidance informatiod,the subsequedk linee of the quatraid can Ue produceO efficiently,which to ensuree that the generateO quatraid收稿日期：2019-W-05；在线发布日期=2020-05-00.本课题得到国家自然科学基金项目上167224,61720）资助.秦川，博士，教授，博士生导师，中国计算机学会上CF）会员，主题研究领域为多媒体信生安全、图像处理等.Ewd：u@bW.Hu.co.王萌，工学学士，主要研究方向为信息隐藏、图像处理•司广文，硕士研究生，主题研究方向为信息隐藏、深度学习•姚恒，博士，副教授，主题研究方向为多媒体取证、信息隐藏、图像处理、模式识别.774计算机学报2021年has an unified theme and there are no obvious semantic deviations between the previous and latter lines.Foe the sama input informatiou,t he model cao generatr multiplr quatraine with the sama themr and meet ths requiremente ob ths patters.By choosing ths different candidats quatraio linee and the information ob keyword,pattere and rhyme,secret information can Vr hidden effectively.Io the experiment,totally25000quatraine werr used no the training datn foe the quatraio generatiou model.Experimental results show that,the hidden secret informatiod cat Vr extracted from thr generaten quatraie with eo error,nd the embeddine ratr ob thr proposed method cae reach aboui 35%,which is sign讦icantlw greatee thae sorne of reported informatioe diding methods based cu text generatioe.We dso utilized the sa_rne set of training datt on oue model with informatioe hiding nd ths curredi typical poetrg generatioe modele withoui informatioe hiding,nd ths trained resulte show W o ths performance indicee of these models,such co perplexitp nd lose value,are close.Then,Vp comparing ths年poeWes randomlp generated with ea_ch of these models,we cad find that their performancee of semantia qualitg foe generated poeWes are W so ^quivaledt.Id Wditicd,the computational complexity，securitp md robustnese of the proposO method are discussed md ths teasibilitg of practical applicatiod is Wso malyzed.Finally,future research direcWm of this worO are gived.Keywords text generatiod；constructive information hidine；quatrains；keywords；embedding rats1引言信息隐藏是将秘密信息隐藏在公开的媒体信息中，使得除收发者外的其他人无法提取出隐藏的信息，甚至不知道有隐藏信息存在的技术.信息隐藏技术作为网络安全的重要研究方向，近年来取得了重大进展，在军事、法律、知识产权保护和隐私保护等领域都有应用年通常秘密信息会被隐藏到各种多媒体载体当中，如图像1音频3视频⑷以及文本等.文本作为使用最为广泛的载体之一，是人们日常沟通交流获取信息的主要媒介，在信息隐藏领域也得到了诸多的应用,具有很高的研究价值.但文本本身信息冗余度较低，因此如何使携带秘密信息的文本接近自然语言文本，并有较高的嵌入率，是文本信息隐藏算法亟待解决的问题％随着信息隐藏技术的发展，通常信息隐藏方案可以划分为两类:载体修改式方案和构造式方案P目前基于文本的信息隐藏方案大部分属于载体修改式，即通过修改载体文本的格式或语法语义进行秘密信息的隐藏.根据修改方式的不同大致又可以将其分为两类.第一类为基于文本格式修改的信息隐藏方案.如通过细微地调整字符间距和行间距来隐藏秘密信息，由于人类视觉系统的特性，很难直接观察出文本的异常4通过调整字符的属性，如磅值、颜色、风格等，也能够隐藏秘密信息如文献[9],通过控制HTML文本中字体的颜色来标记字符，从而映射不同的二进制码，达到秘密传输信息的目的.其它方法如文献9年，通过文档中的不可见字符进行秘密信息的隐藏，在文本每一段结束时键入空白字符,如通过SPACE键和TAB键所生成的不可见字符的长度不同，来隐藏秘密信息.基于文本格式修改的信息隐藏方案不会改变文本的语义，但是对文本格式极为敏感，轻微的格式变化都会导致秘密信息无法正确提取.因此此类方案的鲁棒性不强,同时该类方案统计特征较为明显，容易被隐写分析算法检测出异常.第二类为基于文本语法语义修改的信息隐藏方案.该类方案主要涉及同义词替换方案和句式变换方案9年1总体来说修改文本语法语义的方式,通过阅读感知很难发现这种细微的修改,因此隐蔽性较强.但由于对原始的文本进行了修改和调整，不可避免地会产生失真.在同义词替换方案中载体词汇暴露在文本当中，容易受到隐写分析算法的攻击9年.同时考虑到载体文本的长度有限，可提供替换的词汇也有局限性，因此此类算法的嵌入率一般不高.由于文本修改式信息隐藏算法存在以上的问题,基于文本生成的构造式信息隐藏算法应运而生,并成为当前的研究热点之一.此类算法是根据秘密信息驱动直接生成文本，没有对文本进行任何的修4期秦川等：基于绝句生成的构造式信息隐藏算法774改，因此更加隐蔽安全.文献［5］提出利用生成宋词进行秘密信息隐藏的构造式文本信息隐藏方案.文献［5提据宋词的格式音律要求，选取不同的词汇组成新的宋词文本来隐藏秘密信息，引起了广泛关注.但文献［5提生成宋词的过程中，没有深入考虑文本的语义要求，而是通过韵律规则等进行词语组合，生成的宋词相对缺乏中心思想，容易引起攻击者的怀疑，且该方案的嵌入率也较低.文献5年在文献5提的基础上上行了改进，在嵌入率方面有所提升，但生成文本的质量不高，主题思想不明确，词与词之间的语义匹配仍需要改进.文献5年提出了基于马尔可夫链的宋词生成模型，所生成的宋词质量有所提高，但该方案的嵌入率较低.随着深度学习在机器翻译方面的广泛应用，基于机器翻译模型进行诗词创作的技术日趋成熟.文献［年提出使用循环神经网络(Recurrent Neuim Network,RNA)不断将新生成的诗词压缩到历史向量中，从而生成新的诗词.文献5年提出一种规划模型的诗词生成方案，通过预先规划诗词每一句的主题词，来辅助指导诗词的生成,确保了每一句话的语义信息.规划模型生成诗词的质量较高，但存在主题不明确的问题.文献5年提出了基于工作记忆模型的诗词生成方案.工作记忆模型利用主题信息、历史信息和本地信息，确保生成诗词主题的灵活表达，同时所生成的诗词的质量高，在已报道的文献中处于较好的水平.高质量的诗词生成模型也引起了信息隐藏领域学者的高度关注.文献［20在用规划模型，并通过挑选备选字的方式进行秘密信息的隐藏，获得了较高的嵌入率.但文献［20在过备选词库的组合会有较大的损失，对生成诗词的质量影响较大，导致该方案所生成诗词的质量一般.且随着诗句的生成，由于没有主题信息的限制，可能会丢失主题信息.另一方面，文献［20在秘密信息提取过程需要对模型所能生成的诗词进行遍历，导致提取算法效率较低，无法实现实时提取.针对以上信息隐藏算法所存在问题，本文提出一种基于绝句生成的构造式信息隐藏算法，实现了秘密信息的实时隐藏和提取，且生成绝句诗质量高.绝句属于近体诗的一种形式，由四句组成，有严格的格律要求，常见的绝句有五言绝句和七言绝句.由于绝句文本的信息冗余低，如果强行进行字符的隐藏，必定会造成较高损失，使文本的可读性下降.如何使所隐藏的字符自然地出现在所生成的绝句诗词中是需要解决的关键问题.为此本文提出了基于主题词的高质量绝句生成算法，确保主题词能够自然地出现在所生成的绝句诗词文本中.与文献［20在比，本文算法无需设置内置词库来限制所生成的绝句诗词，从而能够使所生成诗词的损失降到更低，保证生成诗词的可读性.同时我们在训练模型过程中添加了字符的韵律信息，这样生成的诗句抑扬顿挫感强,能够达到自然押韵.在嵌入率方面,我们根据所生成绝句诗的主题词、模板、押韵信息以及备选诗句来进行秘密信息的隐藏，具有较高的嵌入率.实际上，利用本文提出的方案除诗句外还可以对宋词、元曲、对联、歌词等韵律和平仄要求高的文体进行文本生成,进而实现秘密信息的隐藏.需要注意的是，需对不同文体的文本分别进行训练，并采用相应的模板及韵律规则.本文所提出的方案主要在三个方面有所贡献:一是在嵌入率方面有较大提升；二是生成诗词可读性强、韵律自然、主题明确；三是能够实现实时隐藏和提取侗时可对提取的秘密信息进行自我校验，判断文本是否遭受过篡改，安全性更高.相较于基于载体选择的信息隐藏算法，本文所提出的基于文本生成的构造式信息隐藏算法首先在安全性方面表现更好.因为构造式信息隐藏算法没有固定词库，而是通过计算来来行字词的选择和匹配，因此在没有相同训练数据的情况下，很难进行攻击和分析.其次，在生成文本质量方面，构造式信息隐藏算法更加注重字与字之间的搭配,因此生成文本的质量更高.本文第4节介绍所提出算法的整体框架；第0节具体介绍所提出的基于主题词的绝句生成方法;第4节给出信息隐藏算法和提取算法；第4节为实验结果及分析比较;第6节给出全文的总结与进一步的工作.2算法整体框架本文所提出方案的整体框架如图1所示.本文方案主要由三部分组成，分别为模型训练、隐藏过程和提取过程．在模型训练过程中，需要对绝句诗进行预处理,然后根据本文提出算法进行训练，最终生成符合秘密信息隐藏要求的绝句生成模型.收发双方需要共享相关训练参数，确保双方生成的模型达到相同的效果．776计 算 机 学 报2021 年图1方案整体框架图在隐藏过程中，首先根据秘密信息选择对应的 主题词、韵律及模板信息；然后输入模型，生成候选 的绝句诗句；根据秘密信息选择候选诗句，生成含密 绝句诗.在提取过程中，首先根据接收到的绝句诗提取出主题词、韵律和模板信息，然后按照隐藏过程中 生成诗句的方法将相关信息输入模型生成候选诗句；接着根据候选诗句,对照接收到的诗句，提取秘 密信息.以上为本文方案的整体框架,接下来进行详细介绍.3高质量绝句生成模型本文绝句生成模型的基础是基于注意力机制的 序列到序列模型(Sequence to Sequence,Seq2Seq)2 ,因此我们在给出本文的生成模型之前首先对Seq2Seq 模型及工作记忆模型进行相关介绍.3.1 基于注意力机制的Seq2Seq 模型在Seq2Seq 模型中，使用一个循环神经网络作为 编码器，将完整的句子压缩为一个向量.使用另外一个RNN 作为解码器，根据输入向量生成目标句子.但是当输入的句子过长时，这个中间向量难以存储 足够的信息.同时中间向量作为一个整体，无法做到精准地解码，而注意力机制(Sttention Mechanism) 就是为了解决这个问题而设计注意力机制是将当前解码的重点放在当前所对应的字或者片段，而不是整个句子.众所周知冲国古诗词不仅在格式以及韵律上有着严格的要求，同时行与行之间的单个 字或者词都有着紧密的关联.因此，基于注意机制的Seq2Seq 模型在中国古诗词生成过程中有着较好的表现，其生成诗句的基本流程如图1所示.图2诗句生成基本流程4期秦川等：基于绝句生成的构造式信息隐藏算法777这里使用基于注意力机制的Seq2Seq模型，其中编码器以双向LSTMRong Short-Term Memorp）为基础2在编码过程中，第一句诗词输入表示为X=（R,狓，…，狓），输出结果表示为犢=上1，狔'犕）资和n分别代表编码过程中前向传播和反向传播的隐层状态,心表示重置门"表示更新门，y示编码器在第c个字最终输出，计算公式为犱表—tanh(R[2表1•厂表+U狓表上)u n^a(U u hd_1+W u x n g上)E表cr(上表—i+W r x n g(0)表表(1—u-i)•0n-i+u—g dg(4)hg表—n c-n—上)双曲正切函数twh（）为激活函数模以在神经元中引入非线性因素对得神经网络可以逼近任意非线性函数m表的计算过程与nn基本一致.在注意力机制的作用下然码器的第4步解码过程首先计算得出出r，其表示第4步查询状态下第—个字编码输出y y与当前解码结果的相关程度其算公式为_exp（r的仆、a步—帀-------------（6）乂exp（R,的p—年e-=—twh（Us r-\~W a gd g上）其中，其其其是参数矩阵•用G表示第4次解码的结果其表了与解码结果最相关的原始信息其a—与s加权平均其算公式为N步—e步r h i（上d——同时表下一次解码中会作为额外的信息输入到网络当中，S,是解码器在第步解码时的状态.3.2工作记忆模型基本原理在Seq2Seq古诗词生成的基础上，文献［―提出了工作记忆模型.该模型被划分为三个模块，分别为主题信息模块、历史信息模块、本地信息模块.工作记忆模型生成诗词的过程如图3所示.工作记忆模型的基础部分为门控循环单元（Gated Recurrent Unit,GRU）0.在生成过程中，工作记忆模型的输入分为两个阶段，其中第一个阶段的输入为主题词、模板信息和韵律信息；第二个阶段的输入是通过模型计算得到的，包括主题信息、历史信息和本地信息.假设输出诗词由—句组成，表示为｛L ff—;输入主题词共有K个，表示为｛WW—；Mi表示主题信息，分表示历史信息，分表示本地信息分，分，分在表全部工作记忆信息.主题TO厶-2Encoder Encoder Encoder1M］主题信息M2历史信息词晒被全部写入到主题信息犕当中，并且在诗词生成的过程中保持不变，用来指导每一句诗词的生成，确保不偏离主题.在生成第e句诗词L,之前,的每一个字都会被写入本地信息犕中.与其它模型不同，这里并没有将的全部信息写入到M,而是从当中选择最有代表意义的、最贴近主题的词或者字写入到历史信息中，这样可降低对主题的干扰该—为全局追踪向量，记录已经生成的内容，为模型提供全局信息，随着诗句的生成不断更新.为了保证生成诗词主题不偏离，设计了主题追踪机制表I入变量―，以更加明确的方式记录主题的使用信息该为信息的输出，是解码输出，0表示投影参数文代表了第步解码后词汇表中每个字符的概率信息，表达式为s,=GRUR—i年—-1记文，表-年）上）狆上o—1分1”-1，设年）=softmax（R是-（10）式上）中g,代表包含了平仄及押韵的韵律信息.绝句诗词需要满足固定的格式和韵律，因此这里引入韵律分量，用来控制生成绝句诗词的格式和韵律.在训练过程当中定义了36个音韵类别，训练的绝句诗词均为5言绝句或7言绝句.工作记忆模型能够生成高质量的绝句诗词,但是所得到的诗词冗余很低，无法直接进行秘密信息的隐藏.3.3绝句生成模型如何进行秘密信息的隐藏，同时保证生成诗词的质量，是需要解决的问题.通过对工作记忆模型原理的分析,能够得出这样的结论:诗词生成的发起信息主要来源于“主题词”，生成诗词的第一行与主题词的关系最紧密.假设:如果主题词能够出现在生成绝句诗的第一句中，且位置固定这那就可以顺利地将秘密信息提取出来.同时第一行诗词的生成,不依778计算机学报2021年赖于后序诗词，这样可以保证生成诗词的质量不受影响.基于以上分析与假设对们提出了一种基于主题词的高质量绝句生成算法.首先鉴于Seq2Seq模型的诸多优点，文献［18-19,25］利用Seq2Seq模型,分别行行了古诗词生成算法的设计.虽然实施细节有所不同，但是大致可以划分为三个模块.本文以文献［25］为例行行介绍，第一个模块是由词生成诗句(Word-To-Line,WTL)的过程，第二个模块是由诗句生成诗句(Line-LLine对TL)的过程，第三个模块是由上下文生成诗句(Context-To-Line,CTL)的过程.图4是基于Seq2Seq诗词生成过程的示例.在诗词生成的过程当中，首先设置主题词为“秋雁”，然后通过WTL模块，生成第一句诗词；然后以第一句作为输入，经过LTL模块，生成第二句诗词；将第一、二句诗词作为输入，经过CTL模块生成第三句；最终由前三句诗词生成第四句诗词.三个模块的搭建都是以Seq2Seq模型为基础.本文所提出算法点，是对作作记模络作中oLi 部分的改行，也就是由主题词到第一句诗词的生成过程.设主题词的个数为1,在生成第一个字的解码过程中，通过式W可以得出s,的解码状态为s,—1—GRU(so,［32,0,—o提2同2)(18)其中，S。
高一英语大学专业介绍单选题50题(带答案)1.Which major is related to programming and software development?puter ScienceB.EngineeringC.MathematicsD.Biology答案：A。
解析：计算机科学专业与编程和软件开发紧密相关。
工程学专业主要涉及工程设计和建造等方面。
数学专业主要是关于数学理论和计算等。
生物学专业是研究生命现象和生命活动规律的。
2.Which major focuses on building bridges and roads?puter ScienceB.EngineeringC.MathematicsD.Psychology答案：B。
解析：工程学专业会专注于建造桥梁和道路等基础设施建设。
计算机科学专业主要是编程和软件开发等。
数学专业是关于数学理论和计算等。
心理学专业是研究人类心理现象和行为的。
3.Which major requires a strong understanding of algorithms?puter ScienceB.EngineeringC.MusicD.History答案：A。
解析：计算机科学专业需要对算法有很强的理解。
工程学专业主要涉及工程设计和建造等方面。
音乐专业是关于音乐创作和表演等。
历史专业是研究过去的事件和人类社会的发展。
4.Which major is likely to involve designing electronic circuits?puter ScienceB.Electrical EngineeringC.ChemistryD.Literature答案：B。
解析：电气工程专业可能会涉及设计电子电路。
计算机科学专业主要是编程和软件开发等。
化学专业是研究物质的性质、组成和变化等。
文学专业是关于文学作品的创作和研究等。
5.Which major is concerned with the study of human behavior?A.PsychologyB.PhysicsC.MathematicsD.Engineering答案：A。
第２８卷㊀第５期２０２０年５月㊀㊀㊀㊀㊀㊀㊀㊀㊀㊀㊀㊀光学精密工程㊀O p t i c s a n dP r e c i s i o nE n g i n e e r i n g㊀㊀㊀㊀㊀㊀V o l ．２８㊀N o ．５㊀㊀M a y ２０２０㊀㊀收稿日期:２０１９Ｇ１１Ｇ０７;修订日期:２０１９Ｇ１２Ｇ０２．㊀㊀基金项目:国家自然科学基金资助项目(N o ．６１９０１０８１);中央高校基本科研业务费专项资助项目(N o ．３１３２０２０１９９)文章编号㊀１００４Ｇ９２４X (２０２０)０５Ｇ１１６５Ｇ１２基于深度语义分割的多源遥感图像海面溢油监测陈彦彤１,李雨阳１,吕石立２,王俊生１∗(１．大连海事大学信息科学技术学院,辽宁大连１１６０２６;２．交通运输部搜救中心,北京１００７３６)摘要:针对遥感图像海面溢油区域通常受到斑噪声以及强度不均等因素的影响,从而导致溢油区域监测效果较差的问题,本文引入了深度语义分割的方法,将深度卷积神经网络与全连接条件随机场相结合,形成端对端连接.以R e s n e t 结构为基础,首先通过深度卷积神经网络对多源遥感图像粗分割并作为输入,然后经过改进的全连接条件随机场,利用高斯成对势和平均场近似定理,建立条件随机场形成递归神经网络作为输出.通过多源遥感图像对海面溢油区域进行监测,并利用可见光图像估计溢油区域面积.实验在所建立的多源遥感图像数据集上与其它先进模型进行对比,结果表明本文方法提高了溢油区域的分割精度以及精细细节程度,平均交并比为８２．１％,监测效果具有明显地改善.关㊀键㊀词:海面溢油;卷积神经网络;语义分割;条件随机场;遥感图像中图分类号:T P ７５３㊀㊀文献标识码:A㊀㊀d o i :１０．３７８８/O P E ．２０２０２８０５．１１６５R e s e a r c ho no i l s p i l lm o n i t o r i n g ofm u l t i Ｇs o u r c e r e m o t e s e n s i n g i m a g e b a s e d o nd e e p s e m a n t i c s e gm e n t a t i o n C H E N Y a n Ｇt o n g １,L IY u Ｇy a n g １,L ÜS h i Ｇl i ２,WA N GJ u n Ｇs h e n g１∗(１．D e p a r t m e n t o f I n f o r m a t i o nS c i e n c e a n dT e c h n o l o g y ,D a l i a nM a r i t i m eU n i v e r s i t y ,D a l i a n １１６０２６,C h i n a ;２．S e a r c h a n dR e s c u eC e n t e r o f t h eM i n i s t r y o f T r a n s p o r t ,B e i j i n g １００７３６,C h i n a )∗C o r r e s p o n d i n g a u t h o r ,E Ｇm a i l :w a n g js h ＠d l m u ．e d u ．c n A b s t r a c t :I nr e m o t es e n s i n g i m a g e s ,o i ls p i l la r e a s a r e u s u a l l y a f f e c t e d b y s po tn o i s ea n d u n e v e n i n t e n s i t y ,w h i c h l e a d s t o p o o r s e g m e n t a t i o n ．Ad e e p s e m a n t i c s e g m e n t a t i o nm e t h o dw a s i n t r o d u c e d t o c o m b i n e ad e e p co n v o l u t i o nn e u r a l n e t w o r kw i t ha f u l l c o n n e c t i o n c o n d i t i o n a l r a n d o mf i e l d t o f o r ma n e n d Ｇt o Ｇe n d c o n n e c t i o n ．B a s e do n R e s n e t ,f i r s t ,t h e m u l t i Ｇs o u r c er e m o t es e n s i n g i m a g e w a sr o u g h l y s e g m e n t e da s i n p u tb y t h ed e e p c o n v o l u t i o n a ln e u r a ln e t w o r k ．T h e n ,u s i n g G a u s s i a n p a i r w i s ea n d m e a n f i e l d a p p r o x i m a t i o n ,t h e c o n d i t i o n a l r a n d o mf i e l dw a s e s t a b l i s h e da s t h eo u t pu t o f t h e r e c u r r e n t n e u r a l n e t w o r k ．T h e o i l s p i l l a r e ao n t h e s e a s u r f a c ew a sm o n i t o r e db y a m u l t i Ｇs o u r c e r e m o t e s e n s i n gi m a g e a n d e s t i m a t e db y o p t i c a l i m a g e s．E x p e r i m e n t a l r e s u l t s s h o wt h a t t h e p r o p o s e dm e t h o d i m p r o v e s c l a s s i f i c a t i o na c c u r a c y a n dc a p t u r e s f i n e rd e t a i l so f o i l s p i l l a r ea s c o m p a r e dw i t ho t h e rm o d e l su s i n g t h ed a t a s e te s t a b l i s h e db y t h e m u l t iＧs o u r c er e m o t es e n s i n g i m a g e．T h e m e a ni n t e r s e c t i o no v e rt h e u n i o n i s８２．１％,a n d t h em o n i t o r i n g e f f e c t i s s i g n i f i c a n t l y i m p r o v e d．K e y w o r d s:s p i l l e do i lo nt h es e a;C o n v o l u t i o n N e u r a l N e t w o r k(C N N);s e m a n t i cs e g m e n t a t i o n;c o nd i t i o n a l r a n d o mf ie l d;r e m o t e s e n s i n g i m a g e１㊀引㊀言㊀㊀近年来,随着世界各国对原油需求量不断增加,海洋石油运输业迅速发展,但国内外海面溢油污染事故却频繁发生.例如２０１０年,大连新港一艘外籍油轮在卸油时,导致陆地输油管线发生爆炸,事故造成５０平方公里的海域被污染.２０１８年,桑吉轮与货船在长江口发生碰撞[１],事故造成大量石油泄漏溢入东海,对海洋生态环境造成巨大危害.当海面溢油事故发生后,能否准确地检测溢油区域位置和面积信息,对后续采取防治措施至关重要.因此,开展海面溢油监测对于海洋环境保护具有重要意义.传统的海面溢油监测手段为航拍或实地调查,但是这种方法需要投入大量的人力物力,导致成本高㊁操作难度大.而遥感卫星[２]可以不受国界领空的限制,长期且有效地监测海面溢油的情况,是目前海面溢油监测的最佳手段.目前基于遥感卫星的海面溢油监测通常使用星载合成孔径雷达[３](S y n t h e t i cA p e r t u r eR a d a r,S A R),其具有覆盖范围广㊁全天候的优势,能够有效监测海面溢油位置,但S A R图像溢油监测的研究主要集中在探测溢油的有无,对油污面积计算较少.而可见光遥感图像[４]具有分辨率高㊁颜色丰富等特征,可以有效估计溢油区域面积,这对海洋生态破坏评估㊁后续溢油治理起到至关重要的作用.对于多源遥感图像海面溢油监测主要采用图像分割的方法,其中传统的海面溢油分割方法有: (１)基于阈值分割的方法[５],原理是将图像像素点分为若干类.这种方法实现简单且计算量小,但易受海面噪声以及图像灰度分布不均的影响,导致分割准确度低;(２)基于边缘信息的检测方法[６],结合溢油区域的形状特征以及边缘信息,来获取溢油候选区域;(３)基于语义分割[７Ｇ９]的检测方法,它是将图像中属于相同类别的像素聚类为一个区域.可以将溢油区域与海面进行明确的分类,语义分割对图像有更加细致的了解.用于语义分割的传统分类方法有:(１)随机决策森林(R a n d o m D e c i s i o nF o r e s t s,R D F)[１０],利用多棵树对样本进行训练并预测的一种分类方法;(２)马尔科夫随机场(M a r k o v R a n d o m F i e l d s, M R F)[１１],它是一种无向图模型,为每个像素分配定义标记的问题;(３)条件随机场(C o n d i t i o n R a n d o mF i e l d,C R F)[１２Ｇ１３],表示的是给定一组输入随机变量X的条件下另一组输出随机变量Y 的马尔可夫随机场.其中全连接条件随机场(F u l l y C o n n e c t e d C o n d i t i o n R a n d o m F i e l d, F C C R F)克服了传统C R F漏掉细小结构的缺点.但是这些传统方法的分类效果依然较差.近年来,深度学习已经广泛应用于计算机视觉领域[１４],尤其在图像分类方面获得了突破性的成功.L O N GJ等人提出了全卷积网络[１５](F u l l y C o n v o l u t i o n a lN e t w o r k s,F C N),应用于语义分割中,取得了较高的准确率.D e e p l a b[１６Ｇ１７]模型是由G o o g l e团队提出的一种用于语义分割的办法,利用空洞卷积准确调节分辨率,扩大感受野,降低计算量,并用空洞卷积金字塔池化(A t r o u sS p a t i a l P y r a m i dP o o l i n g,A S P P)模块进行多尺度特征提取,得到全局和局部特征.最后用全连接条件随机场,优化边缘效果,解决了因为传统深度卷积神经网络[１８Ｇ２０](D e e p C o n v o l u t i o n a l N e u r a l N e t w o r k,D C N N)中连续的池化和降采样导致空间分辨率明显下降的问题,近一步提升分割效果.但是D e e p l a b模型依然存在一些问题,例如:(１)先用深度卷积神经网络进行粗分割,再利用全连接条件随机场进行细分割,无法实现端对端的训６６１１㊀㊀㊀㊀㊀光学㊀精密工程㊀㊀㊀㊀㊀第２８卷㊀练,导致分类精度低;(２)对海面溢油区域的精细细节提取较差且耗时长.针对以上问题以及多源遥感图像的特点,本文以D e e p l a b 模型为基础,提出一种新的卷积神经网络形式的多源遥感图像海面溢油语义分割模型,用于监测海面溢油区域,该模型将C R F 与深度卷积神经网络相结合,利用高斯成对势和平均场近似定理,建立条件随机场形成递归神经网络(R e c u r r e n tN e u r a lN e t w o r k s ,R N N )[２１],并将它作为神经网络的一部分,获得一个兼具卷积神经网络特性和C R F 特性的深层端对端网络,将其命名为深度语义分割(D e e p S e m a n t i c S e gm e n t a t i o n ,D S S ).利用该模型监测S A R 以及可见光遥感图像的海面溢油区域,并用可见光遥感图像有效估计溢油面积.２㊀D e e pl a b 模型㊀㊀D e e p l a b 模型利用空洞卷积代替了标准的网络卷积操作,通过扩大感受野㊁缩小步幅对特征图采样.它调整了卷积滤波器的感受野来捕捉多尺度的上下文信息,输出不同分辨率的特征.对于一维卷积结构,输出特征图y 的每个位置i ,卷积滤波器w ,输入特征图x ,空洞卷积计算如公式(１)所示.二维卷积结构原理如图１所示.y [i ]＝ðkx [i ＋r k ]w [k ]．(１)图１㊀空洞卷积原理F i g ．１㊀P r i n c i pl e o f a t r o u s c o n v o l u t i o n ㊀㊀首先D e e p l a b 改进了R e s n e t 架构,将R e s n e t 的全连接层转化为卷积层,最后两个池化层去掉了下采样,且后续卷积层的卷积核改为了空洞卷积,并对R e s n e t 的权重进行微调,这样网络最后图２㊀A S P P 模块原理F i g ．２㊀P r i n c i p l e o fA t r o u sS p a t i a l P y r a m i dP o o l i n g(A S P P )７６１１第５期㊀㊀㊀㊀陈彦彤,等:基于深度语义分割的多源遥感图像海面溢油监测输出的特征图分辨率提高,感受野变大.然后进行多尺度提取,A S P P模块如图２所示,在给定的输入特征图上以r＝(６,１２,１８,２４)的３ˑ３空洞卷积并行采样,最后通过A S P P各个空洞卷积分支采样后结果融合到一起,得到最终预测结果.实际上就是通过不同的空洞卷积来对图像进行不同程度的缩放,从而获得了更好的分割效果.最后使用全连接条件随机场细化图片.３㊀海面溢油监测㊀㊀基于多源遥感图像的海面溢油区域普遍存在斑噪声和强度不均的现象.其中S A R图像中许多暗区域会被分类成溢油区域,可见光遥感图像中海杂波等会对目标检测产生影响.首先本文以D e e p l a b模型为基础,输入的图像经过深层卷积神经网络对其进行粗略地分割,然后将全连接条件随机场看做循环神经网络作为输出,再次对图像细分割,实现了深层卷积神经网络与全连接条件随机场端对端的连接,将D C N N与改进的全连接C R F模型优势结合在一个统一的端对端框架中.其次,本文在A S P P模块中加入了B N层,加快网格训练速度.最后,改进了基于平均场近似定理的全连接条件随机场算法,与D C N N端对端连接.３．１㊀全连接条件随机场传统的条件随机场用于平滑噪声,将邻近结点耦合,这样空间上接近的像素会被分配相同标记的标签.但是目前的深度卷积神经网络架构,得分图通常非常平滑.在这种情况下,使用传统的条件随机场模型会漏掉细小结构,目标恢复详细的局部结构.而全连接条件随机场可以克服这个缺点,捕获精细细节.首先将像素标签建模为随机变量,在全局观测条件下形成马尔科夫随机场,设图片为I,x i为像素i的标注,从标签L i中取值,X为随机变量x１,x２, ,x N产生的向量,I 和x的关系可以建模为条件随机场,如公式(２)所示:P(X＝x|I)＝１Z(I)e x p(－E(x|I))．(２)㊀㊀由此可定义吉布斯分布如公式(３)所示.E(x)＝ðiψu(x i)＋ði＜jψp(x i,x j),(３)ψp(x i,x j)＝μ(x i,x j)ðM m＝１w(m)k(m)G(f i,f j),(４)其中:ðiψu(x i)为衡量像素i取标签x i的损失,由深层卷积神经网络获得.ði＜jψp(x i,x j)为衡量像素i,j同时取标签x i,x j的损失,依赖于图像平滑项,使得相似像素更有可能标注相同标签,具体如公式(４)所示.每一个k m G为作用于特征向量的高斯核,f i和f j分别为像素i,j的特征向量,特征向量一般取图像二维位置坐标和颜色R G B向量.距离和颜色的差距越大,则影响越小.m表示高斯核的个数,取１或２.w(m)为权重线性组合,不同卷积核权值不同,它是由网络训练所得到,μ(x i,x j)为兼容性函数,即惩罚项.３．２㊀条件随机场看作循环神经网络根据公式(３),最小化吉布斯分布E(x)就可以得到标注结果,但是过程相对复杂且算法比较耗时,因而本文引入了平均场近似最大后验分布进行推理[２２].Q(x)为条件随机场P(x)的近似,它作为R N N重构.平均场近似推理迭代算法具体步骤如表１所示.表１㊀平均场近似推理迭代算法㊀T a b．１㊀M e a n f i e l da p p r o x i m a t e r e a s o n i n g i t e r a t i v ea l g o r i t h m算法:平均场定理用于将C R F分解为常见的C N N操作Q i(l)ѳ１z i e x p(U i(l))f o r a l l i初始化W h i l en o t c o n v e r g e dd oQ (m)(l)ѳðjʂi k(m)(f i,f j)Q j(l)f o r a l l m消息传递㊅Qi(l)ѳðm w(m)Q (m)i(l)权值调整^Qi(l)ѳðlᶄɪLμ(l,lᶄ)㊅Q i(l)兼容性转换㊅Qi(l)ѳU i(l)－^Q i(l)传递误差微分Q iѳ１z i e x p(㊅Q i(l))归一化E n dw h i l e８６１１㊀㊀㊀㊀㊀光学㊀精密工程㊀㊀㊀㊀㊀第２８卷㊀㊀㊀如表１所示,首先第１步初始化过程为初始概率值,U为卷积神经网络的结果.这个过程相当于在每个像素的所有标签上对一元电位应用一个S o f t m a x函数,不包含任何参数,可以看做神经网络的S o f t m a x层.第２步应用高斯滤波器来实现消息传递,即概率转移.对Q(x)进行高斯模糊,这个过程相当于神经网络的卷积运算.第３步为权值调整,对于每个类标签l,计算出上一步高斯滤波器输出的加权和.可以看作１ˑ１的卷积层,对多个特征图进行卷积运算.第４步为兼容性转换,通过考虑不同标签之前的兼容性并相应的分配惩罚,这样可以获得更好的结果.即如果将不同的标签分配给具有类似属性的像素,这将会受到惩罚,可视为卷积层.第５步用一元势减去兼容性转换步骤的输出,传递误差微分,表示最终概率由卷积神经网络的结果U和全局概率转移结果^Q i(l)共同决定.第６步是归一化操作,将第５步的结果归一化传到下一次循环神经网络迭代作为初始概率.这个过程可以看作另一个没有参数的S o f t m a x操作.本文对平均场近似推理迭代算法的第２步和第３步进行了改进,原始的高斯核为２,考虑了x, y的位置向量和颜色向量,实际上颜色向量在D C N N层中决定了分类的先验概率,所以可以不考虑颜色向量的高斯距离,仅考虑位置差异,这样高斯核为１,距离越远差异就越小.并用全图距离权重网络代替传统的高斯核距离,将距离权重置于全图卷积网络中,该网络的权值由训练样本得到.第２步和第３步概率转移和权值调整合并变为新的算法,相当于卷积运算,如公式(５)所示:Q i(l)＝ðiʂj a i,j Q j(l),(５)其中:a i为距离权重,l为类别,Q j(l)为每个点的类别概率.该算法的迭代过程可表示为多个卷积神经网络层,如图３所示,fθ表示迭代带来的变化.多图３㊀平均场近似推理算法迭代过程㊀㊀F i g．３㊀M e a n f i e l da p p r o x i m a t i o n r e a s o n i n ga l g o r i t h mi t e r a t i v e p r o c e s s层平均场迭代可以重复上述过程实现,每一次迭代都来源于之前迭代的结果,这就相当于将迭代平均场视为一个循环神经网络,网络方程如式(６)~式(８)所示,T为平均场的迭代次数,H１(t)初始值为D C N N归一化的结果㊁H２(t)为一次C R F 过程.Y(t)是第T次迭代的输出,当未达到迭代次数T时则继续迭代.当t＝T时,输出H２(t)为最终迭代结果.H１(t)＝s o f t m a x(U),t＝０H２(t－１),０＜tɤT{,(６) H２(t)＝{fθ(U,H１(t),I),０ɤtɤT,(７) Y(t)＝０,０ɤt＜TH２(t),t＝T{．(８)㊀㊀通过上述改进,模型整体流程图如图４所示,首先输入图片经过R e s n e t网络,将C o n３_x与C o n４_x的中间层改为空洞卷积,然后经过多尺度A S P P模块,通过不同的空洞卷积速率R a t e＝(６,１２,１８,２４)得到不同大小的特征图.其次经过多尺度A S P P模块,加入B N层后,它可以加快训练速度,提高网络的泛化能力.其中A S P P模块的卷积神经网络可视化如图５所示.当感受野较小的时候,提取到的是图像的细节特征.当感受野越来越大时,提取到的是图像的抽象特征.然后经过双线性插值输出特征图,提供了C R F的一元电位,直接以端到端连接到循环神经网络中.最后,进入循环卷积神经网络后,需要迭代t次才能离开循环,使用反向传播算法和随机梯度下降方法进行端到端的培训,一旦离开循环,S o f t m a x层进行网络的终止,输出分类结果.该算法统一了卷积神经网络与C R F的优势,形成了端对端的连接,并细化了分割边缘.９６１１第５期㊀㊀㊀㊀陈彦彤,等:基于深度语义分割的多源遥感图像海面溢油监测图４㊀D S S 流程图F i g．４㊀D S S f l o wc h a rt 图５㊀A S P P 模块的多尺度可视化结果F i g．５㊀M u l t i s c a l e v i s u a l i z a t i o n r e s u l t s o fA S P P ４㊀实验结果与分析㊀㊀实验对基于多源遥感图像的海面溢油区域进行检测,并与其他先进方法进行对比,验证本文模型的优越性.本实验电脑配置为I n t e l i ７处理器,显卡为N V D I AR T X ２０８０T i ,１６G 内存,在L i n u x 系统下的c a f f e 平台,并建立了高质量的S A R 图像与遥感图像数据集,其中可见光遥感图像来源于Q u i c k B i r d ,W o r l d V i e w Ｇ２等遥感卫星所拍摄到的海面溢油图像与G o o gl eE a r t h 中的图像,S A R 图像来源于c 波段雷达s a t Ｇ２极化仪,共采集了４２００张海面溢油图像.所得到的数据集将图像中目标划分为三类,分别为背景㊁溢油区域㊁舰船.然后从图像中随机选择６０％作为训练集,２０％作为验证集,另２０％作为测试集.由于训练需要大量的图片,所以本文进行了数据增强,即对每张图像随机旋转９０ʎ,１８０ʎ,２７０ʎ,最终得到了包含８４００张图像的数据集.实验在训练过程中应用p o l y 学习策略,具体如公式(９)所示.本文将深度卷积神经网络中的迭代次数设为２０K ,批量大小(b a t c hs i z e )为２０,e po c h 表示训练集中全部样本训练一次,共８０次.学习率的初始值为０．００１,为了防止学习率过大,在收敛到最优处会０７１１㊀㊀㊀㊀㊀光学㊀精密工程㊀㊀㊀㊀㊀第２８卷㊀不稳定,所以学习率应随着训练过程按指数级而下降.权重衰减为０．０００５,动量因子为０．９.p o l y ＝１－i t e r n a x _i t e r æèçöø÷po w e r ,(９)其中:p o w e r 为参数,值为０．９,i t e r 表示迭代次数,m a x _i t e r 表示最大迭代次数.本实验包括了海面溢油区域的分类结果㊁m I O U 计算与损失函数㊁时间对比分析以及溢油面积计算.４．１㊀海面溢油分类结果本实验分别对比了遥感可见光图像与S A R 图像的海面溢油分割情况,如下图所示,对比方法分别为C R F ＧR N N ,S P N e t,以及本文所提方法.图６~图８为可见光遥感图像的海面溢油分割结果,可见光遥感图像的油污与海水具有不同的表现.其中图６(a)为原图,可以看出图中溢油区域较明显,但存在海杂波的干扰.图６(b )为C R F ＧR N N 模型检测结果,它只是用传统的卷积神经网络进行粗分割,并且未对平均场定理进行改进,结果将部分海杂波错误分类为溢油区域.图６(c )的S P N e t 模型基本正确分类了溢油区域,但是精确度较差.本文所提出的方法检测结果如图６(d )所示,与图６(b )相比,本文模型可以精确的分类出海杂波与溢油区域,捕获了目标精细细节.图７(a )的溢油区域较分散且不明显,C R F ＧR N N 模型只正确分类了舰船,本文所提出的方法如图７(d )所示,由图可知本文方法分类效果最好,由于实现了端对端的连接分割精度近一步提升.由图８所示,C R F ＧR N N 模型将舰船错误分类为溢油区域,S P N e t 模型与本文模型都正确分类了舰船与溢油区域,但是本文模型对目标的精细分割能力较强,获得了良好的效果.图６㊀可见光遥感图像海面溢油分割结果F i g ．６㊀O p t i c a l r e m o t e s e n s i n g i m a g e o i l s p i l l s e gm e n t a t i o n r e s u l ts 图７㊀可见光遥感图像海面溢油分割结果F i g ．７㊀O p t i c a l r e m o t e s e n s i n g i m a g e o i l s p i l l s e gm e n t a t i o n r e s u l ts 图８㊀可见光遥感图像海面溢油分割结果F i g ．８㊀O p t i c a l r e m o t e s e n s i n g i m a g e o i l s p i l l s e gm e n t a t i o n r e s u l t s １７１１第５期㊀㊀㊀㊀陈彦彤,等:基于深度语义分割的多源遥感图像海面溢油监测图９㊀S A R 图像海面溢油分割结果F i g ．９㊀S A Ri m a g e o i l s p i l l s e gm e n t a t i o n r e s u l ts 图１０㊀S A R 图像海面溢油分割结果F i g ．１０㊀S A Ri m a g e o i l s p i l l s e gm e n t a t i o n r e s u l ts 图１１㊀S A R 图像海面溢油分割结果F i g ．１１㊀S A Ri m a g e o i l s p i l l s e gm e n t a t i o n r e s u l t s ㊀㊀图９~图１１为S A R 图像的海面溢油分类结果.在S A R 图像中,存在固有的相干斑噪声且不２７１１㊀㊀㊀㊀㊀光学㊀精密工程㊀㊀㊀㊀㊀第２８卷㊀同区域特性不均匀,导致分类难度增加,由３幅图可知,C R FＧR N N模型只能正确分类严重溢油区域,即图像中颜色较深的部分.S P N e t模型虽然正确分类了部分溢油区域但是对目标边缘比较模糊,本文所提出的方法分割结果均优于前两种模型.通过实验分析,无论是可见光遥感图像还是S A R图像,本文所提出的方法均适用且分割效果好.４．２㊀m I O U计算以及损失函数将本文模型在所建立的数据集上与其他先进模型进行对比,评价指标为m I O U,称为平均交并比,是语义分割的标准度量,具体如公式(１０)所示.对比结果如表２所示.m I O U值最低的为F C NＧ８s,由于卷积网络的层数较少,从而导致结果较差.D e e p l a b模型的值为７６．５,S P N e t模型的值仅次于本文所提出的模型,达到了７８．９, C R FＧR N N模型m I O U值为７１．８.本文方法与其他先进方法相比效果较好,m I O U达到８２．１. m I O U＝１k＋１ðk i＝０p i iðk j＝０p i j＋ðk j＝０p j i－p i i,(１０)其中:k为类别,i表示真实值,j表示预测值,p i j 表示将类i预测为类j.表２㊀不同模型m I O U对比T a b．２㊀C o m p a r i s o no f d i f f e r e n tm o d e l sm I O UM e t h o d m I O UF C NＧ８s[２３]６０．１D e e p L a bＧM S c[１６]７０．３C R FＧR N N[２２]７１．８D e e p l a b[１６]７６．５H D C[２４]７４．８S P N e t[２５]７８．９HＧR e N e t＋D e n s e C R F[２６]７６．８O x f o r d T V G H OC R F[２７]７７．９P r o p o s e d８２．１㊀㊀端对端模型迭代次数对实验的影响如表３,由表可知,当迭代次数达到５以上,m I O U提高的不是很明显,考虑迭代次数多所占用的时间,所以本文选用迭代次数为T＝５.表３㊀迭代次数对模型的影响T a b．３㊀E f f e c t o f n u m b e r o f i t e r a t i o n s o nm o d e lI t e r a t i o n１２３４５m I O U７９．５８０．０８１．２８１．７８２．１I t e r a t i o n６７８９１０m I O U８２．２８２．３８２．４８２．５８２．６㊀㊀本文利用了交叉熵损失函数(C r o s sE n t r o p y C o s tF u n c t i o n,C E C F)来计算损失,其定义如公式(１１)所示.交叉熵表示真实概率分布与预测概率分布之间的差异,在深度学习中,真实分布已经确定,交叉熵值越小,表示模型预测效果越好.损失函数收敛曲线如图１２所示.H(p,q)＝－ðn i＝１p(x i)l o g(q(x i)),(１１)其中:p(x i)表示真实概率分布,q(x i)表示预测概率分布.图１２㊀损失函数曲线F i g．１２㊀L o s s f u n c t i o n c u r v e４．３㊀时间分析在建立的数据集上与其它先进方法的时间对比,如表４所示.由表可知D e e p l a b模型耗时最长为１．４s,是因为没有实现深度卷积神经网络与全连接条件随机场端对端的连接.耗时最短的模型为F C NＧ８s,而本文所提出的D S S模型耗时相对较短,基本与F C NＧ８s模型处于同一个数量级上且保证了检测精度.３７１１第５期㊀㊀㊀㊀陈彦彤,等:基于深度语义分割的多源遥感图像海面溢油监测表４㊀不同模型耗时分析T a b．４㊀T i m eＧc o n s u m i n g a n a l y s i s o f d i f f e r e n tm o d e l sM e t h o d R u n t i m e/sF C NＧ８s０．５５D e e p l a b１．４S P N e t１．２５C R FＧR N N０．７５P r o p o s e d０．８４．４㊀溢油面积计算海面溢油面积的计算可以为海面污染程度以及未来可能污染的程度提供依据,具有重要的意义.本文通过可见光遥感卫星的分辨率以及溢油区域的像素点数来估计溢油面积,由于可见光遥感图像分辨率高,通常１m分辨率是指一个像素表示地面大约１ˑ１m的面积,因此利用溢油区域像素的数量乘以卫星分辨率的平方,即可求出海面溢油区域面积,具体如公式(１２)所示:S o sʈN o sˑR２,(１２)其中:S o s为海面溢油区域面积,N o s为溢油区域像素点数,R为卫星的分辨率.本文可见光遥感图像的溢油面积如表５所示.表５㊀可见光遥感图像溢油面积T a b．５㊀O i l s p i l l a r e a o f v i s i b l e r e m o t e s e n s i n g i m a g e卫星像素点数卫星分辨率/m溢油面积/m２可见光图像１７８１１２０．６１２９０６５．５可见光图像２２９７７４０．５７４４３．５可见光图像３６１５４８０．５１５３８７５㊀结㊀论㊀㊀本文以深层卷积神经网络为基础,空洞卷积代替最大池化层,并与全连接条件随机场实现端对端的连接,获得一个既有卷积神经网络特性又具有条件随机场特性的深层网络.克服了卫星图像对海面溢油监测的分类较差的问题,提升了捕获目标精细细节的能力.通过S A R和可见光遥感图像对海面溢油区域进行监测,在本文所建立的数据集上m I O U值达到了８２．１,对于遥感图像的分类效果均较好.并且利用可见光遥感图像高分辨率的特点,有效估计了溢油区域的面积,对海洋环境修复及查看海面污染程度具有重要的意义.本文是以弱监督学习为基础的训练模型,网络的训练依赖大量的标记图像,这就需要大量的人力且会受到人为主观因素的影响.因此,未来的研究重点会转向非监督学习,提高算法的便利与可行性.参考文献:[１]㊀童丽珏,周伯煌． 桑吉 轮引发的海洋生态环境损害问题探析[J]．中国环境管理干部学院学报,２０１８,２８(２):１２Ｇ１４．T O N GL Y,Z H O U B H．A n a l y s i so nd a m a g et o t h em a r i n ee c o l o g i c a l e n v i r o n m e n t c a u s e db y＂S a nＧc h i＂o i l t a n k e r[J]．J o u r n a l o f E M C C,２０１８,２８(２):１２Ｇ１４．(i nC h i n e s e)[２]㊀蒋兴伟,何贤强,林明森,等．中国海洋卫星遥感应用进展[J]．海洋学报,２０１９,４１(１０):１１３Ｇ１２４．J I A N GX W,H EX Q,L I N M S,e t a l．．P r o g r e sＧs e so no c e a ns a t e l l i t er e m o t es e n s i n g a p p l i c a t i o ni nC h i n a[J]．A c t a O c e a n o l o g i c a S i n i c a,２０１９,４１(１０):１１３Ｇ１２４．(i nC h i n e s e)[３]㊀康利军,李晓峰,宋甜甜．S A R图像海面溢油检测技术分析[J]．江苏科技信息,２０１８,３５(２０):４１Ｇ４３．K A N G LJ,L IX F,S O N G T T．A n a l y s i so fo i l s p i l ld e t e c t i o n t e c h n i q u e s f o r S A R i m a g e s[J]．J i a n g s u S c i e n c e&T e c h n o l o g y I n f o r m a t i o n,２０１８,３５(２０):４１Ｇ４３．(i nC h i n e s e)[４]㊀李宇,刘雪莹,张洪群,等．基于卷积神经网络的光学遥感图像检索[J]．光学精密工程,２０１８,２６(１):２００Ｇ２０７．L IY,L I U X Y,Z HA N G H Q,e t a l．．O p t i c a l r eＧm o t e s e n s i n g i m a g e r e t r i e v a l b a s e do nc o n v o l u t i o n a l n e u r a l n e t w o r k s[J]．O p t．P r e c i s i o nE n g．,２０１８,２６(１):２００Ｇ２０７．(i nC h i n e s e)[５]㊀A LＧAM R ISS,K A L Y A N K A R N V．I m a g es e gＧ４７１１㊀㊀㊀㊀㊀光学㊀精密工程㊀㊀㊀㊀㊀第２８卷㊀m e n t a t i o nb y u s i n g t h r e s h o l d t e c h n i q u e s[J]．A r XＧi vP r e p r i n tA r X i v:１００５．４０２０,２０１０．[６]㊀赵宏伟,陈霄,刘萍萍,等．视觉显著目标的自适应分割[J]．光学精密工程,２０１３,２１(２):５３１Ｇ５３８．Z HA O H W,C H E NX,L I UPP,e t a l．．A d a p t i v e s e g m e n t a t i o nf o rv i s u a ls a l i e n to b j e c t[J]．O p t．P r e c i s i o nE n g．,２０１３,２１(２):５３１Ｇ５３８．(i n C h iＧn e s e)[７]㊀刘宁波,丁昊,田永华,等．基于组合分形特征的海杂波中目标检测方法[J]．航空兵器,２０１８(２):３８Ｇ４２．L I U NB,D I N G H,T I A NY H,e t a l．．T a r g e t d eＧt e c t i o n m e t h o di n s e ac l u t t e r b a s e d o n c o m b i n e df r a c t a l c h a r a c y e r i s t i c s[J]．A e r o W e a p o n r y,２０１８(２):３８Ｇ４２．(i nC h i n e s e)[８]㊀N O H H,H O N GS,HA N B．L e a r n i n g d e c o n v o l uＧt i o nn e t w o r kf o rs e m a n t i cs e g m e n t a t i o n[C]．P r oＧc e ed i n g s o f t h eI E E Ei n te r n a t i o n a lc o nf e r e n c eo n c o m p u t e r v i s i o n,２０１５:１５２０Ｇ１５２８．[９]㊀C H E N LC,Z HU Y,P A P A N D R E O U G,e t a l．．E n c o d e rＧd e c o d e rw i t ha t r o u ss e p a r a b l ec o n v o l u t i o nf o r s e m a n t i c i m ag es e g m e n t a t i o n[C]．P r o c e e d i n g so f t h e E u r o p e a n C o n f e r e n c eo n C o m p u t e r V i s i o n(E C C V),２０１８:８０１Ｇ８１８．[１０]㊀JS H O T T O N,M J O HN S O N,R C I P O L L A．S eＧm a n t i ct e x t o nf o r e s t sf o r i m a g ec a t e g o r i z a t i o na n ds e g m e n t a t i o n[C]．I E E EC o n f e r e n c e o nC o m p u t e rV i s i o na n dP a t t e r nR e c o g n i t i o n,２００８．[１１]㊀R A N A L L IM,L A G O N AF,P I C O N E M,e t a l．．S e g m e n t a t i o n o fs e ac u r r e n tf i e l d sb y c y l i n d r i c a lh i d d e n M a r k o vm o d e l s:a c o m p o s i t e l i k e l i h o o da pＧp r o a c h[J]．J o u r n a l o f t h eR o y a l S t a t i s t i c a l S o c iＧe t y:S e r i e s C(A p p l i e dS t a t i s t i c s),２０１８,６７(３):５７５Ｇ５９８．[１２]㊀V A R MA KI,K R I S HN AMO O R T H Y S,P I S IＧP A T IRK．N a t u r a l l a n g u a g e q u e r y i n g w i t h c a s c aＧd e dc o n d i t i o n a l r a n d o mf i e l d s[P]．U n i t e dS t a t e sP a t e n tA p p l i c a t i o n２０１２０２５４１４３,２８０,５３５．２０１６Ｇ３Ｇ８．h t t p://w w w．f r e e p a t e n t s o n l i n e．c o m/y２０１２/０２５４１４３．h t m l．[１３]㊀汪萍．基于多尺度条件随机场的语义图像分割深度卷积网络[J]．宿州学院学报,２０１９,３４(７):６９Ｇ７４．WA N G P．D e e p C o n v o l u t i o n a l n e t w o r k w i t hm u l t iＧs c a l e c o n d i t i o n a l r a n d o f i e l d f o r s e m a n t i c i mＧa g e s e g m e n t a t i o n[J]．J o u r n a l o f S u z h o uU n i v e rＧs i t y,２０１９,３４(７):６９Ｇ７４．(i nC h i n e s e)[１４]㊀唐聪,凌永顺,郑科栋,等．基于深度学习的多视窗S S D目标检测方法[J]．红外与激光工程,２０１８,４７(１):３０２Ｇ３１０．T A N G C,L I N G Y S H,Z H E N G K D,e ta l．．O b j e c t d e t e c t i o n m e t h o do fm u l t iＧv i e w S S D b a s e do nd e e p l e a r n i n g[J]．I n f r a r ea n d L a s e r E n g iＧn e e r i n g,２０１８,４７(１):３０２Ｇ３１０．(i nC h i n e s e) [１５]㊀L O N GJ,S H E L HAM E RE,D A R R E L LT．F u l l yc o n v o l u t i o n a ln e t w o r k sf o rs e m a n t i cs e g m e n t a t i o n[C]．P r o c e e d i n g so f t h e I E E E C o n f e r e n c eo nC o m p u t e rV i s i o na n d P a t t e r nR e c o g n i t i o n,２０１５:３４３１Ｇ３４４０．[１６]㊀C H E N LC,P A P A N D R E O U G,K O K K I N O SI,e ta l．．D e e p l a b:S e m a n t i c i m a g e s e g m e n t a t i o nw i t hd e e p c o n v o l u t i o n a ln e t s,a t r o u sc o n v o l u t i o n,a n d f u l l y c o n n e c t e d c r f s[J]．I E E E t r a n s a c t i o n s o np a t t e r na n a l y s i sa n d m a c h i n e i n t e l l i g e n c e,２０１７,４０(４):８３４Ｇ８４８．[１７]㊀C H E N LC,P A P A N D R E O U G,K O K K I N O SI,e ta l．．S e m a n t i ci m a g es e g m e n t a t i o n w i t h d e e pc o n v o l u t i o n a ln e t sa n df u l l y c o n n e c t e dc r f s[J]．a r X i v p r e p r i n t a r X i v:１４１２．７０６２,２０１４．[１８]㊀K R I Z H E V S K Y A,S U T S K E V E RI,H I N T O N G E．I m a g e n e t c l a s s i f i c a t i o nw i t hd e e p c o n v o l u t i o n a ln e u r a l n e t w o r k s[C]．A d v a n c e s i nn e u r a l i n f o rＧm a t i o n p r o c e s s i n g s y s t e m s,２０１２:１０９７Ｇ１１０５．[１９]㊀潘仙张,张石清,郭文平．多模深度卷积神经网络应用于视频表情识别[J]．光学精密工程,２０１９,２７(４):９６３Ｇ９７０．P A N XZ H,Z HA N GS H Q,G U O W P．V i d e oＧb a s e d f ac i a l e x p r e s s i o nr e c o g n i t i o nu s i n g m u l t i m oＧd a l de e p c o n v o l u t i o n a l n e u r a l[J]．O p t．P r e c i s i o nE n g．,２０１９,２７(４):９６３Ｇ９７０．(i nC h i n e s e) [２０]㊀S A I N A T H T N,MO HAM E D A,K I N G S B U R Y B,e t a l．．D e e p c o n v o l u t i o n a l n e u r a l n e t w o r k s f o rL V C S R[C]．２０１３I E E EI n t e r n a t i o n a lC o n f e rＧe n c e o nA c o u s t i c s,S p e e c h a n dS i g n a lP r o c e s s i n g,I E E E,２０１３:８６１４Ｇ８６１８．[２１]㊀G R A V E S A,MO HAM E D A,H I N T O N G．S p e e c hr e c o g n i t i o n w i t h d e e p r e c u r r e n t n e u r a ln e t w o r k s[C]．２０１３I E E EI n t e r n a t i o n a lC o nＧf e r e n c eo n A c o u s t i c s,S p e e c ha n d S ig n a lP r oＧc e s s i n g,I E E E,２０１３:６６４５Ｇ６６４９．[２２]㊀G．L I N,C．S H E N,I．R E I D,A．v a nd a n H e nＧg e l．E f f i c i e n t p i e c e w i s eo fd e e p s t r u c t u r e d m o d e l sf o rs e m a n t i cs eg m e n t a t i o n[J]．A r X i v P r e p r i n tA r X i v:１５０４．０１０１３,２０１５．５７１１第５期㊀㊀㊀㊀陈彦彤,等:基于深度语义分割的多源遥感图像海面溢油监测[２３]㊀D E S MA I S O N A ,B U N E L R ,K O H L IP ,e t a l ．．E f f i c i e n tc o n t i n u o u s r e l a x a t i o n s f o r d e n s e C R F[C ]．E u r o p e a nC o n f e r e n c eo nC o m pu t e rV i s i o n ,S p r i n g e r ,C h a m ,２０１６:８１８Ｇ８３３．[２４]㊀WA N G P ,C H E N P ,Y U A N Y ,e t a l ．．U n d e r Ｇs t a n d i n g c o n v o l u t i o n f o r s e m a n t i c s e g m e n t a t i o n [C ]．２０１８I E E E W i n t e rC o n f e r e n c eo nA p pl i c a Ｇt i o n s o f C o m p u t e r V i s i o n (WA C V ),I E E E ,２０１８:１４５１Ｇ１４６０．[２５]㊀X I A N Y ,C H O U D HU R Y S ,H E Y ,e t a l ．．S e Ｇm a n t i c p r o je c t i o n n e t w o r kf o rz e r o Ｇa n df e w Ｇl a b e l s e m a n t i cs eg m e n t a t i o n [C ]．P r o c e e d i n g so f th e I E E E C o n f e r e n c eo n C o m pu t e r V i s i o na n d P a t Ｇt e r nR e c o gn i t i o n ,２０１９:８２５６Ｇ８２６５．[２６]㊀Z Y A N ,H Z HA N G ,YJ I A ,e t a l ．．C o m b i n i n gt h eb e s t o f c o n v o l u t i o n a l l a y e r sa n dr e c u r r e n t l a y Ｇe r s :A h y b r i dn e t w o r kf o rs e m a n t i cs e gm e n t a t i o n [J ]．A r X i vP r e p r i n tA r X i v :１６０３．０４８７１,２０１６．[２７]㊀A A R N A B ,SJ A Y A S UMA N A ,SZ H E N G ,P ．T o r r ．H i gh e r o r d e r p o t e n t i a l s i ne n d Ｇt o Ｇe n d t r a i n Ｇa b l e c o n d i t i o n a lr a n d o m f i e l d s [J ]．A r X i v P r e Ｇpr i n tA r X i v :１５１１．０８１１９,２０１５．作者简介:㊀陈彦彤(１９８９－),男,辽宁沈阳人,讲师,硕士生导师,２０１２年于吉林大学获得学士学位,２０１７年于中国科学院长春光学精密机械与物理研究所获得博士学位,主要从事遥感图像处理及目标识别方面的研究.E Ｇm a i l :c h e n ya n Ｇt o n g１＠y e a h ．n et ㊀李雨阳(１９９５－),女,辽宁辽阳人,硕士研究生,２０１７年于大连海洋大学获得学士学位,主要从事遥感图像㊁目标识别处理等方面的研究.E Ｇm a i l:l y y７９０８１９＠１６３．c o m ６７１１㊀㊀㊀㊀㊀光学㊀精密工程㊀㊀㊀㊀㊀第２８卷㊀。
Volume 16: The Practice of Psychotherapy000570 General problems of psychotherapy. Principles of practical psychotherapy. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 3-20).Psychotherapy is defined as a kind of dialectic process, a dialogue and discussion between two persons. The various schools of thought on psychotherapy are examined and it is concluded and that their variety does not necessarily invalidate their divergent premises. The interdependence of psyche and body is established as a basic theoretical principle. Initially, the psychotherapist cannot judge the whole of his patient's personality. Although the patient can be seen as approximating the universal man, his individuality is his own and must be allowed expression without being hampered by the doctor's assumptions. Since the individual signifies nothing in comparison with the universal, and the universal signifies nothing in comparison with the individual, methods such as suggestion and 11 mana," the universal healing power, can have some success. This success, however, is believed to be limited by the contradictions of the individual/universal antinomy. Psychoneuroses can be divided into two main groups: collective types with underdeveloped personality and individualists with atrophied collective adaptation. Therapists are cautioned to treat their patients in accordance with the unique and unpredictable individuality of the latter; the cure should not alter the patient's personality but lead to individuation. It is noted that the dialectic procedure calls for the most unbiased attitude possible on the part of the therapist who is a fellow participant in the therapeutic discourse with his patient. Freud's concept of depth dimension logically resulted in the involvement of the doctor's own personality as well as that of the patient in the psychotherapy treatment. Because the conscious attitude of the neurotic must be balanced by compensatory or complementary contents from the unconscious, the continuity of dreams, a source of unconscious content, is stressed. In a 2 month dream series, for example, a water motif, symbolic of the collective unconscious, was recurrent. Another dream series centered around various forms of woman, representing the mythological personification of the unconscious, the "anima." It is justifiable to resort to mythological ideas to assist the patient. Religious beliefs are viewed as forms of psychotherapy which treat and heal the suffering of the soul and the suffering of the body caused by the soul. Since many patients do not have such beliefs, for them, dialectic developments of the mythological material within them is indicated. Various types of people for whom different treatment is required are distinguished. Aim of the therapeutic process is to enable the patient to assimilate the unconscious elements in his psyche, thus achieving the ultimate integration of his personality and the removal of the neurotic dissociation.000571 General problems of psychotherapy. What is psychotherapy. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 21-28).The nature of psychotherapy as a science requiring the use of scientific methods and new forms of treatment is discussed. A glance at former methods shows that suggestion therapy had given way, and that Freud's demands that the causes of neuroses be brought to consciousness were heeded. The trauma theory, originally intended to explain the cause of neuroses is rejected as a hasty generalization, and it is shown that even Freud abandoned this theory for that of repression. It became evident that modern psychotherapy cannot be generally applied but must give undivided and general attention to the individual. This theory goes beyond Freudian psychoanalysis and Adler's individual psychology in that it emphasizes the need for more extensive individualization in the method of treatment. The doctor is cautioned to avoid influencing the patient in the direction of his own bents for only when a man lives in his own way can he be responsible and capable of action. Methods of treatment recommended show that, in all clear cases of neuroses, reeducation and regeneration of personality are essential. It is considered essential that modern psychotherapeutic methods take into account the philosophy of the humane sciences and be based on a thorough knowledge of psychiatry. For adequate study of dreams, a study of primitive psychology, comparative mythology and religion is required.000572 General problems of psychotherapy. Some aspects of modern psychotherapy. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 29-35).Freud's theories concerning the unconscious are examined and evaluated. His assumption that bringing the causes of neuroses into consciousness will cure them is rejected as is his view that dreams are a facade. Instead, dreams are considered to be a reflection of basic elements in an individual's psyche, and regression to childhood is seen as a sign of a patient's efforts to find himself. The negative conception of the unconscious held by Freud is rejected in favor of the belief that the unconscious is a creative factor as well as the stronghold of ancestral conservatism. By discerning the true reasons for his childhood longings, and by penetrating the wide realm of the collective psyche, the patient is enabled to enter into the healthy and natural matrix of the human mind.000573 General problems of psychotherapy. The aims of psychotherapy. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 36-52).Divergent opinions concerning the structure of neuroses and principles of therapy are examined and compared to those derived from clinical observations of patients. Although Freud's explanation of neurosis in terms of sexual causation and Adler's viewpoint based on the will to power are considered to correspond to psychic realities, nevertheless, their methods cannot be applied indiscriminately. Because individual life is extraordinarily diverse, the psychotherapist is exhorted to abandon too fixed an aim and to be aware of the patient's own unconscious nature and will to live. A distinction is made between young patients who must shape their conscious ego and older ones who are more concerned with understanding the meaning of their inner life. Dream analysis is recommended when rational methods of treatment are not successful since, though not scientific, it provides a practical method of showing the patient the trend of his unconscious. It is deemed necessary for the doctor to know as much as possible about primitive psychology, mythology, archaeology and comparative religion in order to enrich the associations of his patients by analogy. Fantasy, the maternally creative side of the masculine mind, is judged to be bound to the taproot of human and animal instinct; therefore, it is considered valid to encourage the patient in his fantasies without going beyond the meaning which is effective for the patient. Painting is encouraged, not for the sake of art, but in order to aid in the development of independence and psychological maturity, freeing the patient from dependence on dreams or the doctor's knowledge. Painting alone is not sufficient, however; it must be accompanied by synthetic interpretation that will allow the patient, in collaboration with the doctor, to attain intellectual and emotional understanding. The problems of dealing with this process of psychic life that lies outside consciousness are stressed; since psychic reality is not commensurate with conscious reality, the observation of the psyche can only be indirect, yet it is this independent psychic reality that the therapist must plumb.000574 General problems of psychotherapy. Problems of modern psychotherapy. In: Jung, C., Collected Works of C, G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 53-75).The use of the term psychoanalysis to describe many different approaches to the study and treatment of the psyche is discussed and alternate therapeutic methods, such as Freudian psychoanalysis, Adlerian individual psychology, medical psychology and analytical psychology are examined. The diversity of professional opinions implied by the various methods indicates the difficulty inherent in gaining access to the psyche. Analytical psychology is defined as embracing both psychoanalysis and individual psychology. This approach includes four stages: confession, elucidation, education and transformation. Each of these stages is subsequently analyzed. In the first stage, the secrets or inhibited emotions, analogous to repressed sins that lead to neuroses, must be confessed to enable the patient to regain his wholeness, and his dependence on the doctor, transference, must be severed. Next the elucidation of this transference must take place and the patient's fixation analyzed. The theories of Freud and Adler are examined in their relation to this stage. The importance of drawing the patient out of himself in aprocess of education in order to attain normal adaptation is stressed. The fourth stage, transformation, isseen to affect both patient and doctor, whose personalities have interacted throughout the treatment. Emphasis is placed on the doctor's need to engage in self-education and to understand that both he andthe patient are in search of a cure that involves not just the body but the entire psyche.000575 General problems of psychotherapy. Psychotherapy and a philosophy of life. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 76-84).The theme that the psyche must be treated with sufficient attention given to physiological and spiritual factors is expanded. The difficulty of the psychotherapeutic task lies in the concept that the psyche, influenced by both physiological and psychological factors, has a contradictory structure giving rise to the problem of opposites. This problem, once brought to consciousness, is seen as capable of threatening the philosophy of life of both therapist and patient. Since man's philosophy of life, the most complex of psychic structures, forms the counterpose to the physiologically conditioned psyche, it ultimately determines one's fate. Hence the psychotherapist needs to have well developed, valid convictions of his own, yet be open to search with the patient for the latter's religious and philosophical ideas. Both religious and philosophical convictions are considered extremely valuable for the psychotherapeutic process, especially the Christian view of original sin and suffering. In an analysis of a patient's neurosis, the psychotherapist is advised to explore both pre-Christian and non-Christian concepts in order to deal with archetypal symbolism in the unconscious. The psychotherapist is enjoined to be aware that the neurosis is not a separate thing but rather the whole of the pathologically disturbed psyche, hence it is not the neurosis but the whole man who should be treated.000576 General problems of psychotherapy. Medicine and psychotherapy. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 84-93).In a 1945 meeting of the Swiss Academy of Medical Science, the differences between medicine and psychotherapy are analyzed. The three stages of medical procedure -- anamnesis, diagnosis and therapy are examined and compared to psychotherapeutic procedures. In psychotherapy, anamnesis, the reconstruction of historical facts in the case, must be supplemented by a consideration of the patient's whole environment. Accordingly the psychotherapist's task involves asking questions seemingly unrelated to the patient's psychoneurosis and applying his professional skill and his intuition to draw out facts about an illness that involves the whole man. In contrast to medical cases, diagnosis is considered irrelevant since in psychotherapy the discovery of facts does not lay the basis for specific treatment but only gives indications of the direction therapy should take; recognition of disease rests much less on the clinical picture than on the content of complexes. Every psychotherapist not only has his own method -- he himself is that method. The subject of psychotherapy, unlike that of medicine, is not the disease but the human being psychically defined. These views lead to the important conclusion that the real and authentic psyche is the unconscious, whereas ego consciousness can be regarded only as an epiphenomenon. The belief that there are certain prenatal modes of behavior and function which bear on the individual psyche, as indicated by the recurrence of common themes in mythology and other folklore, is discussed, and it is determined that neuroses must be considered by the psychotherapist according to their connection with instinctual disturbances and their situational patterns. It is concluded that psychotherapy must go beyond medical symptomatology since it deals with the psyche behind the symptoms and requires knowledge gained from humane sciences, as well as from natural sciences.000577 General problems of psychotherapy. Psychotherapy today. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 94-110).The relationship between the individual psyche and the suprapersonal world of the patient is discussed. This relationship is first experienced within the family but must be broadened to include the entire worldinto which one must enter as a whole person. Projection, the extension of the parental images to thewhole of external reality, is considered to be a major cause of neurosis, and withdrawal of projection of the parental image from external reality one of psychotherapy's most difficult tasks. The recognition ofthe need to facilitate the transition from child to adult is evident in both primitive and Christian ceremonies. The psychotherapist is warned that he must reckon with the patriarchal orientation of the psyche which tradition has encouraged, and also guard against transference whereby the patient develops too great a dependence on the doctor. It is pointed out that the doctor should take into account the patient's religious beliefs in order to achieve the reintegration of the latter's psyche. This process of individuation is seen as the finding of self, a goal recognized in both Yoga and medieval, alchemical, philosophy. The European effort to replace the collectivity of the Church with the collectivity of the State is interpreted as a reflection of the European's deep seated longing for a patriarchal order, for authority. The domination of the individual by the state is rejected because it deprives the former of his rightful destiny. ne aim of psychotherapy is stated: to educate people towards independence and moral freedom. To attempt to subject this aim to a particular political system is viewed as making psychotherapy the tool of totalitarianism. The need to integrate the individual into society is recognized, hence extreme individualism is not advocated. However, it is concluded that the individuation process must give free play to nature and that psychotherapy must be dedicated to the goal of individual development. 2 references.000578 General problems of psychotherapy. Fundamental questions of psychotherapy. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 3 84 p. (p. 111 - 125).The history of man's exploration of the psyche from Paracelsus to Freud is reviewed and the subjective element in theories of psychotherapy is noted and illustrated by reference to Freud's rigid insistence on the pleasure principle and Adler's emphasis on the will to power. In order to avoid too much subjectivity in treating his patients, the doctor is urged to undergo analysis, himself, to gain a knowledge of his own attitudes before treating patients whose attitudes may vary greatly. The psychotherapeutic process is recognized as a dialectic process in which both patient and doctor will become deeply and often painfully involved. Since many factors, social and external as well as internal and unconscious, influence every man and must be taken into account by the doctor, collective traditional values as well as the patient's own individuality arc considered major elements in the reconstruction of his personality. Within each man, there is a superego which may come into conflict with his own psychic system, thereby creating a neurosis. Various types of neuroses are described, and the belief that the patient can be cured, when shown where his conflicts lie, is expressed. To accomplish this cure, each individual's philosophy of life and the compensatory relationship between the conscious and the unconscious must be considered. This implies that the main task of the psychotherapist is to unveil the unconscious without preconceived ideas. Since the unconscious is a collective phenomenon whose existence is indicated by the reappearance of certain archetypes and mythologems throughout the history of man, it is concluded that a true understanding of the unconscious will necessitate a consideration of the unity of the collective unconscious. 2 references.000579 Specific problems of psychotherapy. The therapeutic value of abreaction. In: Jung, C., Collected Works of C, G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 129-139).Abreaction is defined, its therapeutic effect evaluated, and the transference relationship between doctor and patient is discussed. The value of abreaction, i.e. the reenactment of a traumatic experience in achieving the reintegration of the patient's individuality is questioned. Instead, it is proposed that the relationship between doctor and patient has the greatest therapeutic effect: it provides the patient with the support needed to overcome the effects of his neuroses; it requires the doctor to enter into the psyche of his patient probing deeply into the origins of the neuroses in order to lay the groundwork for thepatient's return to reality. The strictly sexual interpretation of the patient's neurosis is regarded as limitedin its value. Transference, the dependence of the patient on the doctor, is seen as a stage that will lead to the patient's attaining a position of equality with the doctor. It is considered important that the patientsee not only the nature and cause of his neuroses but also envisage his own psychological goal, his own growth toward a whole psyche.000580 Specific problems of psychotherapy. The practical use of dream-analysis. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 139-161).The rejection of dream analysis by many as a valid method to penetrate the unconscious is acknowledged, but its practical usefulness is upheld. Several cases are cited to support the conclusion that dreams must be taken as diagnostically valid facts. Dreams are often anticipatory and provide valuable information for understanding the patient and planning his treatment. Although the doctor himself may regard his patient's dreams as confused, he must realize that his confusion ties in his own lack of understanding, and that a true understanding will be attained when both doctor and patient reach agreement on the dream content as the result of joint reflection. Suggestion is considered to have serious limitations as a therapeutic technique, since it can hamper the patient's willingness to face his own problems and to make his, own judgments and decisions. The analysis of dreams is preferred because the nocturnal unconscious psychic activity, covers a wide gamut of conditions and developments. They require careful interpretation by the doctor; however, nothing in them is certain but uncertainty. For an adequate understanding, a series of dreams must be analyzed rather than a single dream taken out of context. The idea of sublimation proposed by Freud to save man from the imaginary claws of the unconscious is rejected since the unconscious is not considered to be any more demoniacal than the conscious. Compensation -- the balancing of elements in both the conscious and the unconscious -- is considered to be a basic law of psychic behavior. Unconscious compensation is only effective when it cooperates with an integral consciousness; however, the values of the conscious personality must remain intact. The belief that the patient's philosophical, religious and moral convictions must be taken into account is reiterated. The conclusion is drawn that dreams reveal more clearly the evolutionary stratification of the psyche than does the conscious mind and that assimilation of unconscious contents can lead the conscious mind back to the natural state of the patient's own being, 1 reference.000581 Specific problems of psychotherapy. The psychology of the transference. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 163-201).The problem of transference is reviewed and discussed in terms of alchemical symbolism in order to provide some orientation in the study of the unconscious. Mystic marriage, coniunctio, which played an important part in alchemy and was concerned both with chemistry and a kind of mystic philosophy, is discussed, since it sheds light on the mystery of chemical combination and is related to transference in psychology. The history of this symbolism is outlined because of its usefulness in the exploration of the unconscious. The complexity of transference, a partly instinctive process difficult to weigh, is analyzed, and the danger inherent in taking over the patient's psychic sufferings is mentioned. The incestuous aspect in transference is recognized and is described as a flurry of unleashed opposites within the awakened unconscious that must be reconciled. These assaults of the unconscious are symbolized in alchemy by Mercurius who is both friend and panacea, good and evil. Religions, particularly Christianity, are considered to be great psychotherapeutic systems. Modern man's hostility to religion is considered as increasing the danger of dissociation between the ego conscious and the unconscious. The many analogies that exist between alchemical symbols and the study of the unconscious show that the study of this medieval speculative philosophy and science can shed great light on the process of individuation and the analysis of the unconscious. 18 references.000582 Specific problems of psychotherapy. The psychology of the transference: an account of thetransference phenomena based on the illustrations to the "Rosarium Philosophorum." 1. The mercurial fountain. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 203-210).The symbols used in the first illustration of the "Rosarium Philosophorum," an illustration of the mercurial fountain, are described and analyzed in terms of the psychology of the unconscious. The illustration contains the alchemical symbols of quaternity -- the four elements which merge into a fifth entity, the quinta essentia, the circular basin, and the seven planets. This last symbol evokes the image of Mercurius, the mother of all seven, since he is also his own mother and father. The three essences of Mercurius are represented by the serpent. The sun and moon appear above the triad as the indispensable parents of the mystic transformation. The progression from 4 to 3 to 2 to 1, the alchemical axiom of Maria, represents four mutually antagonistic tendencies that must be reduced to a final unity. The quaternity, one of the most widespread archetypes, represents the arrangement functions by which the conscious mind takes its bearings; the circle expresses the primordial image of man and the soul, and the number four stands for the pluralistic state of a man who has not yet attained inner unity. The triad, the agent of upwelling, appears as masculine; the dyad, as feminine. Psychologically the triad represents desire, instinct and determination, while the dyad corresponds to the psyche's reaction to the conscious mind's decision. The water signifying the unconscious, represents the ebbing and flowing of the circle of life. These symbolic forms of universal archetypes are considered to be unconscious projections of the alchemists. 3 references.000583 Specific problems of psychotherapy. The psychology of the transference: an account of the transference phenomena based on the illustrations to the "Rosarium Philosophorum." 2. King and queen. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 211-235).The alchemic king and queen figures, symbolizing both the marital and the incestuous brother/sister relationship, are discussed and interpreted in the light of the psychological processes of projection and transference. The gesture of the clasped left hands, pointing to a secret path, is seen as a reference to man's dark, unconscious side and to the dubious nature of the affective relationship: heavenly and earthly as well as incestuous. The right hands, holding flowers, appear compensatory. The 4/5 arrangement of the flowers represent fire and air as the active, masculine element, water and earth as the passive, feminine element with the fifth flower as the quinta essentia. Although the symbols in this illustration all emphasize nature, a hidden, spiritual meaning is also suggested as is the triple nature of Mercurius -- masculine, feminine and divine - who was seen as the earth-bound partner of the Holy Ghost, who wards against the ominous darkness of nature's secrets. Mercurius is beyond sex, in a state reached only when male and female come together into one. The concept of Anthropos is viewed as an eternal idea, present in various cultures. The love factor in the illustration, a love with a sinister touch of incest, is analyzed, and the conclusion is drawn that incest symbolizes union with one's own being -- individuation or becoming a self. By means of several diagrams the complex nature of projection and transference that occur in the doctor/patient psychotherapeutic relationship is analyzed. The doctor's task is to make the patient conscious of the meaning of the transference in order to enable him to arrive at the ultimate reintegration of his personality. The counter crossing of the sexes symbolized in the Rosarium and in fairy tales is illustrated and analyzed. This counter crossing occurs in various and complex forms in life. On the psychological level, the tangle of cross-cousin marriage is seen in transference. In a world of mass mindedness, it is deemed necessary that the individual consciously attempt to integrate his personality in order to counter the psychic dissociation of modern man. 17 references.000584 Specific problems of psychotherapy. The psychology of the transference: an account of the transference phenomena based on the illustrations to the "Rosarium Philosophorum." 3. The naked truth. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press,1966. 384 p. (p. 236-240).The illustration of the nude couple, symbolic of Sol and Luna, is analyzed as a symbol of the unionbetween the conscious mind and the shadow, The accompanying text indicates that the alchemists recognized the moral and psychological implications of their opus. The presence of the dove above thenude figures indicates that this union is spiritual as well as erotic. The illustration is analyzed psychologically: instinct and the primitive psyche 4ave emerged into consciousness; ego and shadow are no longer divided; a precarious unity has been attained. 1 reference.000585 Specific problems of psychotherapy. The psychology of the transference: an account of the transference phenomena based on the illustrations to the "Rosarium Philosophorum." 4. Immersion in the bath. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press. 1966. 384 p. (p. 241-246).Several versions of the descent into the bath motif are presented and this theme is discussed in terms of its religious and psychological implications. According to the alchemists, this "solutio," or dissolution of the king and queen, represents a return to the dark initial stage in the mother's womb; within the lustral water, a new birth is being prepared. The king stands for the spirit; the queen represents the body. The two can only be united in the soul, a hermaphroditic being symbolized by the mixture of oil and water.A connection is made between this alchemical representation and the Mass, wherein the two substances are mixed in the chalice. The immersion in the bath is interpreted psychologically as a descent into the unconscious, the dove represents the union of the opposites from above and the bath represents the union from below: opposition and identity. When interpreted in the light of transference, it represents the stage at which the sexual fantasies coloring the transference are recognized while the longing for transcendental wholeness remains. 9 references.000586 Specific problems of psychotherapy. The psychology of the transference: an account of the transference phenomena based on the illustrations to the "Rosarium Philosophorum." 5. The conjunction. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 247-256).The illustration of the king and queen immersed in water and joined in a passionate embrace is analyzed as a biological symbol for the union of opposites. The coniunctio brings to birth something that is one and united. The water is seen as the boiling solution in which the two opposing substances unite. In psychology, the transference, as symbolized by the coniunctio, comes about unintentionally. In trying to resolve the transference, the doctor is cautioned to avoid doing violence to the patient's unconscious. 2 references.000587 Specific problems of psychotherapy. The psychology of the transference: an account of the transference phenomena based on the illustrations to the "Rosarium Philosophorum." 6. Death. In: Jung, C., Collected Works of C. G. Jung, Vol. 16. 2nd ed., Princeton University Press, 1966. 384 p. (p. 257-266).The alchemical illustration of the king and queen, lying dead in the sea that has become a sarcophagus, is interpreted as an archetypal image of a stage in the individuation process. The king and queen have melted into a single being with two heads. Once the opposites unite, all energy ceases. The picture represents putrefactio -- the decay of a once living person from which new life will arise, hermaphroditus. There is a hint of incest which must be punished. The coniunctio is seen as the union of lucid consciousness and nocturnal light, the unconscious, of the active, masculine side of woman -- animus -and the passive, feminine side of man -- anima. Animus and anima represent semi-collective。
基于有向图卷积与门控循环单元的短时交通流预测方法*崔文岳1谷远利1赵胜利2芮小平3▲（1.北京交通大学综合交通运输大数据应用技术交通运输行业重点实验室北京100044；2.山西太行云顶文化旅游发展有限公司山西晋城048000；3.河海大学地球科学与工程学院南京211100）摘要：为了充分挖掘快速路交通流时空特性，解决当前城市快速路交通流预测存在交通流时空特性挖掘不充分等问题，进一步提高城市快速路短时交通流的预测精度与效率，研究了基于有向图卷积神经网络和门控循环单元的组合模型（directed graph convolution network-gate recurrent unit，DGC-GRU）的城市快速路短时交通流预测方法。
该方法提出空间相关性矩阵并将其引入图卷积神经网络中，构建有向图卷积神经网络用于表征交通流的有向性和流动性。
将交通流参数输入有向图卷积神经网络后得到有向图卷积算子，并将有向图卷积算子引入门控循环单元，通过有向图卷积神经网络捕捉交通流的空间特性，通过门控循环单元捕捉交通流的时间特性，输出快速路交通流预测结果。
选取西雅图环形快速路感应器检测数据进行实例分析，对比模型预测效果。
结果表明：在数据集与参数设置均相同的情况下，DGC-GRU交通流预测模型的训练收敛速度更快，且平均绝对误差（mean absolute error，MAE）和平均绝对百分比误差（mean absolute percentage error，MAPE）均优于对比模型，与传统的GRU、GCN、DGC-LSTM模型相比，DGC-GRU模型能够将MAE和MAPE指数分别降低33.01%、5.76%、1.32%和27.75%、1.15%、7.76%，表明DGC-GRU交通流预测模型能够有效挖掘城市快速路网中的交通流时空分布特征，具有良好的预测精度与效率。
关键词：城市快速路；短时交通流预测；深度学习；图卷积神经网络；门控循环单元中图分类号：U491.1+4文献标识码：A doi：10.3963/j.jssn.1674-4861.2023.02.013A Method of Predicting Short-term Traffic Flows Based on aDGC-GRU ModelCUI Wenyue1GU Yuanli1ZHAO Shengli2RUI Xiaoping3▲(1.Key Laboratory of Transport Industry of Big Data Application Technologies for Comprehensive Transport,BeijingJiaotong University,Beijing100044,China;2.Shanxi Taihangyunding Cultural Tourism Development Corporation,Jincheng048000,Shanxi,China;3.School of Earth Sciences and Engineering,Hohai University,Nanjing211000,China)Abstract：In order to study the spatiotemporal characteristics of traffic flows on urban expressways which have not been fully explored in previous studies,a short-term prediction method for traffic flow on urban expressways is pro-posed,in order to improve the prediction accuracy and efficiency based on a combined model of directed graph con-volutional neural networks and gated recurrent units(DGC-GRU).The proposed method uses a spatial correlation matrix,which is also combined with a graph convolutional neural network.A directed graph convolutional neural network(DG-CNN)is developed to characterize the directionality and variability of traffic flows.Traffic flow pa-rameters are input into the DG-CNN to obtain the directed graph convolution operator,and the directed graph convo-收稿日期：2022-02-22*国家自然科学基金项目（41771478）、北京市科技计划项目（Z121100*********）资助第一作者简介：崔文岳（1998—），硕士研究生.研究方向：交通规划与管理.E-mail:*******************▲通信作者：芮小平（1975—），博士，教授.研究方向：交通地理信息系统理论与应用研究等.E-mail:***************0引言精准可靠的城市快速路短时交通流预测是城市交通控制与交通诱导的关键保障，同时也成为城市智能交通系统中不可或缺的环节。
强化学习强化学习笔记(一)1 强化学习概述Alpha Go 的成功，强化学习（Reinforcement Learning，RL）成为了当下机器学习中最热门的研究领域之一。
与常见的监督学习和非监督学习不同，强化学习强调智能体（agent）与环境（environment）的交互，交互过程中智能体需要根据自身所处的状态（state）选择接下来采取的动作（action），执行动作后，智能体会进入下一个状态，同时从环境中得到这次状态转移的奖励（reward）。
强化学习的目标就是从智能体与环境的交互过程中获取信息，学出状态与动作之间的映射，指导智能体根据状态做出最佳决策，最大化获得的奖励。
2 强化学习要素强化学习通常使用马尔科夫决策过程（Markov Decision Process，MDP）来描述。
MDP数学上通常表示为五元组的形式，分别是状态集合，动作集合，状态转移函数，奖励函数以及折扣因子。
近些年有研究工作将强化学习应用到更为复杂的MDP形式，如部分可观察马尔科夫决策过程（Partially ObservableMarkov Decision Process，POMDP），参数化动作马尔科夫决策过程（Parameterized Action Markov Decision Process，PAMDP）以及随机博弈（Stochastic Game，SG）。
状态（S）:一个任务中可以有很多个状态，且我们设每个状态在时间上是等距的；动作（A）:针对每一个状态，应该有至少1个操作可选；奖励（R）:针对每一个状态，环境会在下一个状态直接给予一个数值回馈，这个值越高，说明该状态越值得青睐；策略（π）:给定一个状态，经过π的处理，总是能产生唯一一个操作a，即a=π(s),π可以是个查询表，也可以是个函数；3 强化学习的算法分类强化学习的算法分类众多，比较常见的算法有马尔科夫决策过程算法（MDP），Q-Learning算法等。
University life is a significant phase in a students journey,offering a blend of academic challenges,personal growth,and social experiences.Heres an insight into what university life might entail,written in an English essay format.Title:The Essence of University LifeIntroduction:The transition from high school to university is a major step in a young adults life.It is a time of exploration,learning,and selfdiscovery.University life is not just about attending classes and studying it is a multifaceted experience that shapes an individuals future. Academic Pursuits:At the heart of university life is the pursuit of knowledge.Students engage in a rigorous academic curriculum designed to challenge their intellect and broaden their horizons. Lectures,seminars,and practical sessions provide a platform for students to delve deeper into their chosen field of study.The academic environment encourages critical thinking, research,and innovation.Personal Development:Beyond the classroom,university life offers ample opportunities for personal development.Students are encouraged to step out of their comfort zones,take on leadership roles,and participate in various clubs and societies.This not only helps in honing their skills but also in building their character.Social Interactions:One of the most enriching aspects of university life is the social interaction it offers. Students from diverse backgrounds come together,creating a vibrant and multicultural community.Friendships forged in these years often last a lifetime,providing a support network that extends beyond the campus.Campus Life:The campus itself is a hub of activity.From sports events to cultural festivals,there is always something happening.These activities not only provide a break from the academic rigor but also offer a chance to showcase talents and interests outside the classroom.Living Arrangements:For many students,university life means living away from home for the first time.This experience teaches independence and selfreliance.Dormitories or shared housing arrangements provide a unique living experience,where students learn to coexist andrespect each others space and privacy.Career Preparation:Universities also serve as a bridge to the professional world.Internships,workshops,and career guidance sessions are integral parts of university life,preparing students for the job working events and industry interactions provide a glimpse into potential career paths.Conclusion:In essence,university life is a crucible where students are not only educated but also nurtured into becoming wellrounded individuals.It is a period of immense growth,both intellectually and personally,laying the foundation for a successful and fulfilling future.。
