Non-Local Sparse Unmixing for Hyperspectral Remote Sensing Imagery
Yanfei Zhong,Member,IEEE,Ruyi Feng,and Liangpei Zhang,Senior Member,IEEE
Abstract—Sparse unmixing is a promising approach that acts as a semi-supervised unmixing strategy by assuming that the observed image signatures can be expressed in the form of linear combinations of a number of pure spectral signatures that are known in advance.However,conventional sparse unmixing in-volves?nding the optimal subset of signatures for the observed data in a very large standard spectral library,without considering the spatial information.In this paper,a new sparse unmixing algorithm based on non-local means,namely non-local sparse unmixing(NLSU),is proposed to perform the unmixing task for hyperspectral remote sensing imagery.In NLSU,the non-local means method,as a regularizer for sparse unmixing,is used to ex-ploit the similar patterns and structures in the abundance image. The NLSU algorithm based on the sparse spectral unmixing model can improve the spectral unmixing accuracy by incorporating the non-local spatial information by means of a weighting average for all the pixels in the abundance image.Five experiments with three simulated and two real hyperspectral images were performed to evaluate the performance of the proposed algorithm in compar-ison to the previous sparse unmixing methods:sparse unmixing via variable splitting and augmented Lagrangian(SUnSAL)and sparse unmixing via variable splitting augmented Lagrangian and total variation(SUnSAL-TV).The experimental results demon-strate that NLSU outperforms the other algorithms,with a better spectral unmixing accuracy,and is an effective spectral unmixing algorithm for hyperspectral remote sensing imagery.
Index Terms—Hyperspectral remote sensing,non-local,sparse unmixing,spatial information,spectral unmixing.
I.I NTRODUCTION
M IXED pixels are often encountered in remote sensing imagery,due to the sensors’insuf?cient spatial resolu-tion and the spatial complexity.In order to deal with the problem of spectral mixing and effectively identify the components of the mixed spectra in each pixel,the spectral unmixing technique was proposed,which estimates the fractional abundances of the pure spectral signatures or endmembers in each mixed pixel[1]. There are two basic models used to analyze the mixed pixel problem:the linear mixture model(LMM)and the nonlinear
Manuscript received January31,2013;revised April18,2013;accepted Au-gust19,2013.Date of publication September13,2013;date of current version August01,2014.This work was supported by the National Natural Science Foundation of China under Grant41371344and the Foundation for the Author of National Excellent Doctoral Dissertation of P.R.China(FANEDD)under Grant201052.(Corresponding author:Yanfei Zhong.)
The authors are with the State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University,Wuhan430079, China(e-mail:zhongyanfei@https://www.doczj.com/doc/756628821.html,;fry1988@https://www.doczj.com/doc/756628821.html,;zlp62@whu. https://www.doczj.com/doc/756628821.html,).
Color versions of one or more of the?gures in this paper are available online at https://www.doczj.com/doc/756628821.html,.
Digital Object Identi?er10.1109/JSTARS.2013.2280063mixture https://www.doczj.com/doc/756628821.html,pared with the nonlinear mixture model, which is based on a nonlinear mixing assumption inside the re-spective pixel[2],the LMM expresses the measured spectrum of each mixed pixel in any given spectral band as a linear com-bination of the endmembers,with relative concentrations at the respective spectral bands.The LMM holds when the mixing scale is macroscopic,and it has been widely applied for many different applications,due to its computational tractability and ?exibility.
The traditional linear spectral unmixing algorithm consists of an endmember extraction or generation step,and an abun-dance estimation step.A variety of endmember extraction al-gorithms have been proposed to?nd the most spectrally pure signatures for the input image,including the pixel purity index [3],N-FINDR[4],orthogonal subspace projection[5],and the hybrid endmember extraction algorithms[6].The endmember generation algorithms,such as iterative error analysis[7]and iterative constrained endmembers(ICE)[8],are based on the assumption that pure signatures are not present in the input data. However,these algorithms are very likely to fail in highly mixed scenarios[2].
Sparse unmixing,as a semi-supervised method,is based on the assumption that the observed image can be expressed in the form of linear combinations of a number of pure spectral signatures from a large spectral library that is known in advance [9].Because the size of the spectral library is often large,the number of endmembers in the spectral library will be much greater than the number of spectral bands.The sparse unmixing model is a typical underdetermined linear system,in that it is dif?cult to?nd a unique,stable,and optimal solution.To solve the problem,the sparse unmixing algorithm via variable splitting and augmented Lagrangian(SUnSAL),as one of the representa-tive algorithms,was proposed by Bioucas-Dias and Figueiredo [2].In SUnSAL,the spectral library is built a priori,and it uses the alternating direction method of multipliers(ADMM)to ef-?ciently solve the constrained sparse regression[10].However, the SUnSAL approach just treats the remote sensing image as a group of digital signals,without considering any of the spatial information that the image may possess.To integrate the spatial information,sparse unmixing via variable splitting augmented Lagrangian and total variation(SUnSAL-TV)was developed, which utilizes the spatial information between each pixel and its neighbors in the sparse unmixing formulation by means of a TV regularizer.SUnSAL-TV has obtained better spectral unmixing results than SUnSAL and the traditional unmixing techniques [11];however,it just accounts for the spatial homogeneity of the ?rst-order pixel neighborhood system of the abundance map, and cannot make full use of other potential spatial information.
1939-1404?2013IEEE.Personal use is permitted,but republication/redistribution requires IEEE permission.
See https://www.doczj.com/doc/756628821.html,/publications_standards/publications/rights/index.html for more information.
In this paper,a new sparse unmixing method based on non-local means,namely non-local sparse unmixing(NLSU), is proposed by utilizing all possible self-predictions in the abundance maps.In NLSU,a non-local regularizer based on the non-local means method is added to incorporate the non-local spatial information into the primary sparse unmixing formulation,based on the basic sparse unmixing model.The non-local means method estimates the value of the current pixel as an average of the values of all the pixels whose Gaussian neighborhood is similar to the neighborhood of the current one. This approach can effectively preserve the spatial correlation among image features,and takes advantage of the redundancy and self-similarity in an image[12].The non-local means algorithm has been widely used in image denoising[13],noise reduction for hyperspectral imagery[14],noise estimation for hyperspectral imagery denoising[15],image segmenta-tion[16],super-resolution reconstruction[17],hyperspectral imagery restoration[18],image classi?cation[19],and so on.In the proposed NLSU algorithm,the non-local means is used as a non-local regularizer.Unlike the total variation(TV) regularizer in SUnSAL-TV,which accounts for the spatial homogeneity of the?rst-order pixel neighborhood system,the non-local means regularizer can utilize all possible self-pre-dictions about the similar sparsity distribution in the estimated abundance image,and extracts the spatial information by averaging the set of pixels in a certain size of sliding window of the image[12].Based on the non-local idea,as well as the advantages of the sparse unmixing algorithms,the proposed NLSU algorithm can avoid the need to estimate the number of endmembers and extract the endmembers from the observed image.In addition,compared with SUnSAL-TV,which uses the spatial information in a limited local window,NLSU takes into account the non-local spatial information of the whole abundance image.
The proposed method was tested and compared with the other sparse unmixing algorithms,SUnSAL and SUnSAL-TV,using three simulated and two real hyperspectral images.The exper-imental results demonstrate that the proposed NLSU algorithm can obtain a smoother abundance map and achieve a better spec-tral unmixing accuracy.
The rest of this paper is organized as follows.In Section II, we review sparse unmixing theory.The NLSU algorithm is de-scribed in detail in Section III.Section IV presents a descrip-tion of the datasets used and analyzes the experimental results. Section V discusses the sensitivity of NLSU in relation to the different parameters.Conclusions are drawn in Section VI.The Appendix introduces the process of how to solve the proposed model in detail,and acts as an extension of Section III.
II.S PARSE U NMIXING
A.The Linear Mixture Model(LMM)
The LMM has been widely used to determine and quantify the abundance of materials in mixed pixels;it assumes that the spectral response of a pixel in any given spectral band is a linear combination of all of the endmembers present in the pixel.For example,in a typical remote sensing scenario,a mixed pixel can be represented as,where
is the number of the spectral bands.Meanwhile,suppose
is a spectral signature(or endmember)matrix,which is denoted as,where is a1column vector standing for the th endmember,and is the number of endmembers existing in the observed image.Then,let
be a1abundance vector,where denotes the fraction of the th endmember present in pixel. The spectral signature of the pixel vector can be represented by the LMM as follows:
(1)
where is a1vector denoting the noise and model error. Considering the ground truth,we suggest that two constraints are imposed on the LMM:the abundance non-negativity con-straint(ANC)and the abundance sum-to-one constraint(ASC), as follows:
(2)
(3)
B.Sparse Unmixing
Sparse unmixing?nds a linear combination of endmembers for each observed pixel from a large spectral library,as shown in Fig.1(a)–(b).The“sparse land model”for hyperspectral un-mixing,which is derived from sparse and redundant representa-tion theory[20],?nds the best combination of standard spectral signatures(columns with different colors)in the large spectral library,with the optimal abundance for the observed pixel.
As shown in Fig.1(a),most of the abundance values are zeros (the white ones),and there are also empty abundance maps in the abundance matrix in Fig.1(b),which illustrates the sparsity of the abundance distribution.In addition,it is critical to prop-erly build the spectral https://www.doczj.com/doc/756628821.html,pared with the conventional LMM,the sparse unmixing model can be rewritten as
(4)
where acts as the available spectral library,which is a large matrix,and denotes the abundance vector corresponding to library.,is the number of bands,and is the number of endmembers in.Due to the small number of endmembers contributing to a mixed pixel,is sparse,which makes the sparsity constraint useful.If is treated as the total observed image which satis?es,and is the number of pixels in the observed image,then achieved from and.
is extremely sparse,due to most of the lines being full of zeros. The estimated abundance maps can then be output by setting a small threshold(e.g.,)to tell which lines in are the valid values rather than zeros,and the number of the endmembers and the signatures of the endmembers can be obtained.
Mutual coherence[21]–[24],an important indicator,must be mentioned when discussing sparse unmixing,especially for a redundant spectral library.Mutual coherence is used to mea-sure the degree of coherence for the endmembers of the spectral
ZHONG et al.:NON-LOCAL SPARSE UNMIXING FOR HYPERSPECTRAL REMOTE SENSING IMAGERY
1891
Fig.1.The sparse unmixing model.(a)Sparse land model for hyperspectral unmixing.(b)Sparse unmixing.
library matrix.The calculation method for mutual coherence is outlined below,and its value ranges from zero to one.Many studies have suggested that the coherence has a strong in?u-ence on the uniqueness of the sparse solution.Unfortunately, the spectra of the endmembers are highly correlated,and the mutual coherence of the library will be close to one.Due to the small number of endmembers present in a mixed pixel[2],the poor coherence situation can be mitigated,to some extent,by the sparse nature of the mixed pixels’spectral combination.
(5) where is a column vector denoting the th endmember in, and stands for the norm.is the mutual coherence of library.
Bearing the LMM and sparse unmixing theory in mind,the sparse unmixing problem based on the LMM for each mixed pixel of the hyperspectral image can be written as
(6) where is the abundance vector,and denotes the number of nonzero components of.Just like,the noise and model error in LMM,,is the tolerated error derived from the noise or the model itself.However,(6)contains the term,which is a typical NP-hard problem,and it was dif?-cult to solve until Candès and Tao[25],[26]proved that the norm can be replaced by the norm under a certain condition of the restricted isometric property(RIP).Hence,the previous problem can be restated as a convex optimization problem,as follows:
(7) where,and is the th value of.The above constrained optimization problem can be converted into an un-constrained version by minimizing the respective Lagrangian function,as shown:
(8)
where parameter is a non-negative parameter which controls the relative weight of the sparsity of the solution.The terms and represent the ANC and ASC,respec-tively.The is de?ned in the set,which obeys rules such as when,otherwise.In order to compute the objective criterion,many different strate-gies have been proposed,such as orthogonal matching pursuit (OMP)[27],basic pursuit(BP)[28],and iterative spectral mix-ture analysis(ISMA)[29].
Classical sparse unmixing can solve the spectral unmixing problem,but no spatial information is taken into account. SUnSAL-TV brought the spatial contextual relationships into the sparse unmixing model,and has led to better unmixing re-sults.The model is shown in(9)–(10)[11].Meanwhile,the total
1892IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING,VOL.7,NO.6,JUNE2014
variation(TV)regularizer works for the spatial information by promoting piecewise smooth transitions in the abundance maps,aiming at weakening the oscillatory patterns among the ?rst-order neighborhood pixels.
(9) where
(10)
is denoted as a series of the neighboring pixels of in abundance matrix.Furthermore,and are column vectors of the abundance matrix.repre-sents the set of horizontal and vertical neighbors in.
repre-sents the norm of.
as denotes the th column of.
III.N ON-L OCAL S PARSE U NMIXING FOR H YPERSPECTRAL
R EMOTE S ENSING I MAGERY
Differing from SUnSAL-TV,in this paper,a new sparse un-mixing method that utilizes the spatial information,namely the non-local sparse unmixing(NLSU)algorithm,is proposed for hyperspectral remote sensing imagery.NLSU,which is based on the non-local means method,combines with the non-local spatial information and makes systematic use of all possible self-predictions that the abundance maps can provide.
A.The Non-Local Means Method
Unlike the TV regularizer,the non-local means method deals with each pixel by replacement based on the information of the whole image.The motivation behind the development of NLSU for hyperspectral remote sensing imagery was to exploit the similar patterns and structures in an image,because of the high degree of redundancy in each image,which means that every small window in an image will have many other similar win-dows in the same image.The non-local means algorithm has been widely used in hyperspectral image processing[14],[15], [18],[19]to provide the non-local spatial information or spatial constraint.
Just as two neighboring pixels will have similar abundances for the same endmember,every small window in the abundance map will have many similar windows in the same map or image. The non-local means method estimates the abundance of one pixel as an average of the abundances of all the pixels whose Gaussian neighborhood is similar to the neighborhood of the current one,as shown in(11)[12],
[13]:
(11)Fig.2.The principle of the non-local means model.
where denotes the abundance matrix,and stands for the location of one pixel in the abundance image, or we call it the label of the pixel,as does.
is the normalizing factor.Accordingly,,,and are the neighborhoods or the similarity windows of pixels,,and.is a Gaussian kernel with standard deviation,and acts as a decay pa-rameter.Fig.2describes the principle of the non-local means model.In Fig.2,we assume is one searching window of the abundance image,which is used to restrict the number of pixels taken into account in the weighted average,and is the current similarity window centered at pixel.The non-local means operator estimates the abundance of as an average of the abundances of all the pixels whose Gaussian neighborhoods show the same features as the current one(e.g., the similarity window,),and the weight equals, ,where,act as the basis of the similarity measure.For more speci?c details about non-local means, please refer to[12],[13].
B.Non-Local Sparse Unmixing
Based on the non-local means method and sparse unmixing model in(8),the non-local sparse unmixing model is built up as follows.Join the non-local means term,the abundance sum-to-one and non-negative constraint terms,and rewrite the optimization problem.The model of the proposed NLSU algorithm is as follows:
(12) The gradient-based function concerning both the non-local structures and textures is
(13)
ZHONG et al.:NON-LOCAL SPARSE UNMIXING FOR HYPERSPECTRAL REMOTE SENSING IMAGERY1893
where with denotes one line of the abundance matrix,and,in reality,the line vector can represent one piece of the abundance map corresponding to one kind of end-member,i.e.,the th endmember here.is a non-local gra-dient which is de?ned as the vector of all the partial deriva-tives at pixel.and are two abundance values located at and in the th endmember’s abundance map,respectively.is a weight used to measure the abun-dance similarity between and.
For the last two terms,the computations are
(14)
(15) For the whole objective function,we give some further ex-planation below.The spectral library obtained in advance is ex-pressed as(is the number of endmembers in the library),is the set of abundance maps,
is the observed data,is the number of bands,and is the total number of pixels of the observed data.and are the spar-sity constraint parameter and the non-local spatial constraint pa-rameter,respectively.The?rst term,computed as,is the data term,which represents the?delity of the estimated abundances.de-notes the norm,and represents the non-local means operator for the estimated abundance maps.In(13), ,,,and is a real function:
,which means that the pixels in a two-dimensional abun-dance map correspond to a set of one-dimensional abundance values.is based on the non-local gradient de?ned above as,and it is utilized to measure the differences in the multiple directions around in the searching window.In ad-dition,is also the norm of the weighted graph gradient.Together with the last two terms,and are denoted as the ANC and ASC,respectively,and an indicator function of a non-empty closed convex set, i.e.,
if
if
In NLSU,,,de?ned on a two-dimensional discrete grid.The non local gradient is calculated as follows:
(16) where is a positive measure de?ned between pixel and.To keep with standard notations related to graphs,the weight written as,taking the place of the above measure ,is de?ned as
(17)
Thus,is used to measure the abundance similarity be-tween pixels and,and satis?es the conditions
and.The weights are de?ned as
(18) where is the normalizing constant:
(19)
A neighborhood system on the abundance map,, is a family of subsets of.denotes a square neighborhood of pixel with a?xed size,and the similarity be-tween two pixels and depends on the Euclidean distance of the abundance vectors and,which both locate in the same searching window.Furthermore,the weight is asso-ciated with the Euclidean distance of these two abundance vec-tors.is the standard deviation of the Gaussian kernel(), and acts as the degree of?ltering,which controls the decay of the exponential function.rep-resents the Gaussian weighted Euclidean distance.
Based on the NLSU model in(12),the main implementation framework of NLSU is as follows.
1)Step1:Building the Standard Spectral Library:The spec-tral library is collected by?eld spectrometer or from other ex-isting standard spectral libraries,such as the USGS spectral li-brary or ASTER library.To minimize the mismatch between the spectral library and the observed hyperspectral remote sensing imagery,calibration is essential during the data preparation step.
2)Step2:Constructing and Solving the NLSU Model:Based on the classical LMM,the sparse unmixing objective function is formulated.Considering the non-local spatial information in the sparse unmixing model,the non-local means operator,acting as a regularizer,is added to the sparse unmixing model,as shown in(12).To solve(12),ADMM is adopted to decompose the dif?cult problem of(12)into a sequence of simpler ones.The process of obtaining the?nal optimal solution of the unmixing results,,as well as each intermediate variable,(,,, ,,,,,,,,),is detailed in the Appendix.
3)Step3:Stopping Condition:As the optimization problem needs several loop iterations,if the stopping criterion is satis-?ed,the NLSU algorithm computing procedure has been com-pleted,and the unmixing results can be output.Otherwise, ,and the procedure should be iterated again.The stopping condition for computing the?nal optimal(i.e.,the abundance matrix)is to satisfy the following inequality:
where is a small constant,and,,,, ,,,represent the intermediate results of the th iteration.
The?owchart for NLSU is shown in Fig.3.
1894IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING,VOL.7,NO.6,JUNE
2014
Fig.3.The?owchart of NLSU.
IV.E XPERIMENTS AND A NALYSIS
The NLSU algorithm was coded in MATLAB7.8.0and was tested on three groups of simulated hyperspectral datasets and two real hyperspectral images.Consistent comparisons between the NLSU algorithm and the classical sparse unmixing algo-rithms of SUnSAL[2]and SUnSAL-TV[11]were also com-pleted.The accuracy assessment of all the experiments in this paper was made by computing the signal-to-reconstruction error (SRE)[2].
A.Simulated Datasets
In this experiment,a total of342spectral signatures were ran-domly selected from the USGS spectral library to build the ini-tial spectral library.The mutual coherence of the library was very close to1.In order to verify the stability of the proposed method,we chose two groups of a?xed number of spectral sig-natures whose spectral angles(SA)were larger than7degrees, which can be regarded as quite different ones,as well as a group whose SA were all smaller than4degrees,which means they can be easily confused.For simulated datasets constructed from the standard spectral library,and utilized by the library,the cal-ibration step is omitted.
Simulated data1was generated following the method-ology of[31],with7575pixels and224bands per pixel,using two groups of?ve randomly selected spectral signatures from library,denoted splib06 (https://www.doczj.com/doc/756628821.html,/spectral.lib06).This dataset obeys the LMM,using the?ve randomly selected signatures as the endmembers,and the ASC was imposed in each simulated pixel.In the generated image,illustrated in Fig.4(a),there are pure regions as well as mixed regions,constructed using mixtures ranging between two and?ve endmembers, distributed spatially in the form of distinct square regions. Fig.4(b)–(f)shows the true abundances for each of the?ve endmembers.The background pixels were made up of a mixture of the same?ve endmembers,and their respective fractional abundance values were?xed to0.1149,0.0742, 0.2003,0.2055,and0.4051.Finally,Gaussian noise was added with SNR30dB.
Unlike simulated data1,data2was generated by a nonlinear model Euclidean distance function,with8080pixels and224 bands per pixel,using two groups of ten randomly selected spec-tral signatures from library.We also imposed the ASC in this dataset and got a series of abundance images with gradual changes in spatial distribution.The data were also contaminated by Gaussian noise(SNR30dB).Fig.5(a)–(j)shows the true abundance maps of the ten endmembers.
The fractional abundances of the endmembers of simu-lated data3(Fig.6)follow a Dirichlet distribution uniformly over the probability simplex[9],[32],and were provided by Dr.M.D.Iordache and Prof.J.M.Bioucas-Dias[11],[31].This dataset exhibits a very good spatial homogeneity,which can reveal the spatial features quite well for the different unmixing algorithms.Similarly,it was also contaminated with the same level of Gaussian noise,with SNR30dB.
In this section,we test the performance of the proposed NLSU algorithm using the three simulated datasets,and Figs.7–12show the unmixing results using NLSU,SUnSAL, and SUnSAL-TV,respectively.Here,we just show part of the estimated endmembers’abundances,and the signatures corresponding with the abundance distribution in the standard spectral library.Due to the use of the same standard spectral library for all the sparse unmixing processes,the endmember signatures are exactly consistent with the different algorithms. Assessments are made from both the qualitative and quantita-tive aspects,and the parameters are set as shown in Table I. Generally speaking,when SA7,the estimated abundances are more accurate and have a better visual effect,because it is much easier to tell the different materials https://www.doczj.com/doc/756628821.html,pared with Fig.8,the results in Fig.7are far better,and are closer to the true abundances.In Fig.8,the background of endmember5, which was obtained by SUnSAL,is full of noise points,and it is dif?cult to make out the endmember signature from the mixed spectra without considering any spatial information.The reason for this is that the spectral angle distances are quite small,which makes it dif?cult to separate the endmembers from noise,and it also re?ects the limitations of the traditional sparse unmixing method.
Throughout the estimated abundance results,from Figs.7–12,the algorithms considering spatial information achieve a better visual effect,have smoother spatial changes in the homogeneous regions,and contain fewer noise points. Like SUnSAL-TV,NLSU also shows the potential of a spatial term in the procedure of sparse unmixing.However,due to the different spatial operators,the observable effects in the results are not the same.Overall,NLSU gets a better spatial consistency,both in?at zones such as the background,and the edges of regions,due to the non-local means operator,which
ZHONG et al.:NON-LOCAL SPARSE UNMIXING FOR HYPERSPECTRAL REMOTE SENSING IMAGERY
1895
Fig.4.True fractional abundances of simulated data 1.(a)The simulated image.(b)The true abundance of endmember 1.(c)The true abundance of endmember 2.(d)The true abundance of endmember 3.(e)The true abundance of endmember 4.(f)The true abundance of endmember
5.
Fig.5.True fractional abundances of simulated data 2.(a)Endmember 1.(b)Endmember 2.(c)Endmember 3.(d)Endmember 4.(e)Endmember 5.(f)End-member 6.(g)Endmember 7.(h)Endmember 8.(i)Endmember 9.(j)Endmember 10.
decides the current pixel’s abundance by matching the patterns of similarity windows that have the same sparse distribution in the whole search window.In contrast,the use of the TV regular-izer in SUnSAL-TV considers a ?rst-order pixel neighborhood system and is based on the fundamental hypothesis that two neighboring pixels will have similar fractional abundances for the same endmember [11].It can be observed from Fig.9that the results of NLSU have an overall uniformity,but sometimes may show a little block effect (e.g.,endmember 5in Fig.12).Unfortunately,the abundance images obtained by SUnSAL-TV may exhibit an over-smooth visual effect around some pixels,even under an optimal combination of regularization parame-ters,and some of the structures and details are modi ?ed to a certain extent (e.g.,the transition regions in endmember 9in Fig.9and endmember 10in Fig.10);even some straight edges are not well preserved (e.g.,the edges in Fig.8).
From the experimental results,it can be seen that the ac-curacy of the results increases after considering the spatial
1896IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING,VOL.7,NO.6,JUNE
2014
Fig.6.True fractional abundances of simulated data3.(a)True abundance of endmember1.(b)True abundance of endmember2.(c)True abundance of end-member3.(d)True abundance of endmember4.(e)True abundance of endmember5.(f)True abundance of endmember6.(g)True abundance of endmember7.
(h)True abundance of endmember8.(i)True abundance of endmember9.
information,whether by TV regularizer or non-local means operator.It therefore makes sense to apply the idea of incor-porating the spatial smoothing operator properly,to get rid of small,wrong unmixing units and maintain most of the correct information.Furthermore,compared with TV regularization, the proposed method can obtain a higher accuracy,both in the quantitative analysis and in the visual effect.Meanwhile,the statistics in Table II also con?rm the potential of the proposed method from the quantitative point of view.The SRE values of SUnSAL-TV here are obtained by a non-isotropic type of total variation,which gets better results in most cases.
Table II gives a comparison of SUnSAL,SUnSAL-TV,and the NLSU methods in terms of SRE under the conditions of two different SA values(one larger than7degrees,and the other smaller than4degrees),for the three groups of simulated data. NLSU improves the SRE from14.8867to28.3598for data2 when SA is larger than7degrees,and it also improves the SRE from15.1477to29.674for data1when SA is larger than7de-grees,with the greatest improvement being as high as14.5266.However,SUnSAL-TV is a little inferior when compared with NLSU in these three groups of simulated data.
Table II also shows the running times of each algorithm, which are closely related to the numbers of iterations(for more about this,see Table I about parameter settings),as well as the programming environment.In these experiments,the core process of NLSU was coded in Visual C++6.0and MATLAB 7.8.0,while the others were all programmed in the MATLAB platform.Furthermore,the use of certain“clever”tactics can also help to obtain the optimal ef?ciency,such as applying a fast Fourier transform(FFT)to the image and then analyzing the FFT data in the frequency domain,as in SUnSAL-TV, which can greatly improve the ef?ciency.
B.Real Datasets
The?rst hyperspectral dataset used in the real data ex-periments was collected from the Cuprite mining district in west-central Nevada in1997.The size of the test area we chose was250191,with188bands(see Fig.13(a)–(b)).
ZHONG et al.:NON-LOCAL SPARSE UNMIXING FOR HYPERSPECTRAL REMOTE SENSING IMAGERY
1897
Fig.7.Estimated abundances of endmembers1,2,and5for simulated data1,using the different algorithms when.
The standard spectral library for this data is the USGS library containing498pure endmember signatures.Essential cali-bration was undertaken in order to mitigate the mismatches between the hyperspectral image and the signatures in the li-brary[2].
The5050subset of the second hyperspectral dataset (Nuance data),with46bands(Fig.14(a)),was taken by the Nuance NIR imaging spectrometer(650–1100nm and10nm spectral interval)in April2012.The spectral library shown for this dataset,containing52pure materials(Fig.14(e)),was acquired from other Nuance datasets which were also obtained by the Nuance NIR imaging spectrometer at the same time.The possible calibration mismatches between the observed hyper-spectral data and the spectral library were handled according to[2],[11].To undertake a quantitative assessment,we also took the same scene on the same day using a digital camera with150150pixels(HR),for its high spatial resolution(see Fig.14(b)).The geometrical calibration for LR and HR,acting as a preprocessing,is an indispensable part in building a better simulation of the true abundance distribution.We?rst selected one band of the LR hyperspectral image as the base image, and then registration was made between the base image and the HR image by ENVI software.After geometrical calibration for the Nuance data and the HR image,we undertook classi-?cation to HR by support vector machine(SVM),which was implemented by ENVI software(Fig.14(c)shows the ground truth selected manually as ROIs).The HR classi?cation result was down-sampled to obtain the approximate true abundance images(the LR abundance image),given that the scale factor between the HR image and the Nuance data was3[33].The re-sult of the classi?cation is shown in Fig.14(d),which includes three major land-cover classes of dead leaves,fresh grass, and background.The approximate true abundance images are shown in Fig.15.
The parameter settings for all the algorithms are listed in Table III.We then display all the unmixing abundance im-ages using the different algorithms,as shown in Figs.16–17. Table IV shows the unmixing accuracy of the Nuance dataset with the different methods,as it has an HR image for the reference.
1898IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING,VOL.7,NO.6,JUNE
2014
Fig.8.Estimated abundances of endmembers1,2,and5for simulated data1,using the different algorithms when.
The visual comparisons of the three sparse unmixing ap-proaches in Fig.16show the varying degrees of unmixing ac-curacy for the two typical minerals,alunite and buddingtonite. All three unmixing methods obtain similar unmixing results on the whole.However,compared with SUnSAL,the algorithms with spatial consideration re?ect better spatial consistency of the minerals.NLSU also keeps the essential shape corners. For the buddingtonite,the power of spatial regularization is evident,even for the widely distributed https://www.doczj.com/doc/756628821.html,pared with SUnSAL-TV,the spatial extraction or elimination force of NLSU is much stronger,and it is hard to say which one is better,as the intensity needs to be at just the right level. However,generally speaking,NLSU can capture much more concise information about the object of interest.From Table IV, the computational time costs for the different methods with the Cuprite data are all signi?cant,because of the complexity and the large size of the data.
A visual comparison of the abundance images achieved with the real hyperspectral data2suggests that the proposed method is more successful in keeping a better consistency of surface features than the other three approaches.For component3in Fig.17,the abundance images achieved by NLSU are smoother and preserve more complete spatial relationships,compared with the classical SUnSAL method.Taking the component2, fresh grass,as an example in real data2,it can be seen that the transitions are much sharper in the abundance maps obtained by SUnSAL,while the SUnSAL-TV and NLSU images are much more?at.Furthermore,the results of NLSU are more precise than SUnSAL-TV,especially in the area of left bottom. This con?rms that unmixing incorporating non-local spatial information can?nd and contain more detailed information lying in the same region.
Table V provides us with more evidence of the advantage of NLSU in the experiments using the Nuance data.For the Nu-ance data,the SRE value of the SUnSAL algorithm is3.819. However,the SRE increases to6.0015after taking the non-local spatial information into consideration.In addition,the algorithm of SUnSAL-TV,which adopts total variation of the?rst-order pixel neighborhood system as the spatial regularization,also achieves a higher accuracy than SUnSAL.The proposed method gets an improvement of over2when compared with SUnSAL, and exhibits good potential for the characterization of mixed
ZHONG et al.:NON-LOCAL SPARSE UNMIXING FOR HYPERSPECTRAL REMOTE SENSING IMAGERY
1899
Fig.9.Estimated abundances of endmembers5,9,and10for simulated data2,using the different algorithms when.
pixels via a non-local method.For the computation time,due to the large number of iterations,NLSU costs more time than the other two algorithms.On the whole,the proposed algo-rithm outperforms the traditional sparse unmixing methods,and shows good stability with these general real hyperspectral re-mote sensing images.
V.S ENSITIVITY A NALYSIS OF NLSU
As with SUnSAL and SUnSAL-TV,NLSU has more than three parameters in the objective criteria that may in?uence the performance of the algorithm.The parameters consist of the two regularization parameters(and)in(12)and some other latent parameters in the spatial terms,the non-local means terms(the radius of the similarity window,the ra-dius of the searching window,and the decay coef?cient ),and other parameters,including the default iterations,de-fault tolerations,and the regularization parameter.Among them,some are empirical values,such as and,together with and,which can achieve better results after a few adjustments,and they can remain unchanged for any dataset.However,some parameters,especially,,,and, which have a great impact on the objective criteria’s optimiza-tion,or are sensitive to different data and signi?cantly in?uence the unmixing accuracy,need to be adjusted until they reach an ideal effect.These parameters vary from image to image,de-pending on the features of the image,e.g.,the information den-sity,and they greatly in?uence the https://www.doczj.com/doc/756628821.html,st but not least, whether to consider the ASC is also of interest.To analyze the effects of setting these parameters,and the impact of the ASC when running the NLSU algorithm,simulated data1(SA 7),as shown in Fig.4,was used to undertake unmixing with different values of the parameters,and under different physical constraints.
A.Impact of the Regularization Parameters
In the NLSU model presented in(12),the regularization pa-rameters and play signi?cant roles in controlling the relative contribution of the data?delity,sparse constraint,and spatial relationships,as well as the physical meaning.To ana-lyze the impact of the regularization parameters when running the NLSU algorithm,we plotted the curve of the SRE values
1900IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING,VOL.7,NO.6,JUNE
2014
Fig.10.Estimated abundances of endmembers5,9,and10for simulated data2,using the different algorithms when.
with a series of different combinations of and,as shown in Fig.18.
was assumed to have the following values:
and was assumed to be
It can be seen that a better unmixing quality is acquired when the value of reaches,and is less than. The larger the selected,the lower the SRE obtained.It can also be con?rmed that the proportion of the non-local operator is much lower than the sparsity constraint term in simulated data1 (SA7),which needs to enhance the regularization param-eter while using a small penalty()for the sparsity term. In order to achieve a preferable sparsity as well as exploit the useful spatial information,appropriate regularization parame-ters should be found so that the relative contribution can better suit the optimization.
B.Sensitivity to the Non-Local Means Operator Parameters The non-local means operator,as the spatial term,contains several parameters,some of which are empirical values,as men-tioned previously,while some of the other parameters signi?-cantly in?uence the unmixing quality.and are two latent parameters in the non-local operator,and have important im-pacts on the whole algorithm.and were analyzed as fol-lows.was assumed to have the following values:
The relationship between the value and the un-mixing accuracy(SRE)with is presented in Fig.19(a).was assumed to have the following values:
. The relationship between the value and the unmixing accu-racy(SRE)with is presented in Fig.19(b).
ZHONG et al.:NON-LOCAL SPARSE UNMIXING FOR HYPERSPECTRAL REMOTE SENSING IMAGERY
1901
Fig.11.Estimated abundances of endmembers2,5,and9for simulated data3,using the different algorithms when.
As shown in Fig.19(a),the more appropriate the value, the higher the SRE.The reason for this is as follows:An ap-propriate value makes the best split and a more accurate intermediate,e.g.,,,,can be achieved,and the unmixing results are very close to the abundance truth.
As shown in Fig.19(b),lower or higher values result in a smaller SRE,but the lower values lead to a faster decrease in SRE.When the value is set in the middle of the value range, SRE is higher.The reason for this is as follows:A lower value results in weaker spatial information extraction,with much noise and wrong unmixing pixels remaining,while a much higher value leads to stronger spatial regularization,which can result in the loss of abundant structural or gradient variation information,making the spatial information over-smooth.Both situations are bad for the unmixing procedure.
C.Impact of the Abundance Sum-to-One Constraint(ASC)on the Sparse Unmixing Algorithm
To be physically meaningful,the ANC was applied in all the models above.However,the ASC of each mixed pixel was not enforced on every original optimization criteria.In order to verify the impact of the ASC on the sparse unmixing algorithms,the results obtained by the four different models, SUnSAL,SUnSAL-TV,NLSU(without ASC),and NLSU (with ASC),with optimal parameter settings(listed as Table I, simulated data1,SA7),were compared.Fig.20shows the different impacts of the ASC on the sparse unmixing.
It is illustrated that adding the ASC to the NLSU algorithm brings about a better SRE(29.67433)for the unmixing opti-mization than SUnSAL(15.1477)and SUnSAL-TV(25.8333), as well as NLSU(without ASC)(25.9162).As the unmixing is used for the purpose of material quanti?cation,both the ASC and ANC should meet the requirements of the physical principle at the same time,which re?ects the true abundance fractions of the materials.It can be seen from the results presented in Fig.20 that enforcing the ASC on the NLSU algorithm gives a better SRE.
VI.C ONCLUSION
To improve the accuracy of spectral unmixing,this paper pro-poses a non-local sparse unmixing(NLSU)algorithm which can
1902IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING,VOL.7,NO.6,JUNE
2014
Fig.12.Estimated abundances of endmembers 2,5,and 9for simulated data 3,using the different algorithms when .
TABLE I
P ARAMETER S ETTINGS W ITH THE S IMULATED D
ATA
incorporate more spatial information and simultaneously pre-serve the structural information.As the non-local means method can take advantage of the redundancy and self-similarity in an image,and keep ?ne structure,details,and texture,NLSU adopts the approach of estimating the value of each abundance vector as an average of all the pixels whose Gaussian neighbor-hood is similar to the neighborhood of the current pixel,after sparse unmixing without any spatial consideration.
ZHONG et al.:NON-LOCAL SPARSE UNMIXING FOR HYPERSPECTRAL REMOTE SENSING IMAGERY 1903
TABLE II
A C OMPARISON OF THE T HREE S PARSE U NMIXING M
ETHODS
Fig.13.Cuprite data.(a)Cuprite data.(b).Spectral library.
Three simulated hyperspectral datasets and two groups of real hyperspectral remote sensing images were used to test the performance of the proposed NLSU.The experimental results in this paper consistently show that the NLSU algorithm provides better results than the traditional SUnSAL,and can achieve results that are close to or even better than the results of the latest SUnSAL-TV algorithm.Due to the non-local method,the NLSU algorithm can consider spatial information more globally for the abundance maps.The sensitivity analysis of the
parameters
,,,and demonstrates the importance of choosing the proper parameter values to guarantee the accuracy of the proposed algorithm.Our future work will focus on the building of the spectral library,as it is quite signi ?cant for the sparse unmixing methods.We will also look to improve the ef ?ciency of the proposed algorithm,as it is currently quite complex.In addition,we plan to further enhance our spatial consideration approach based on non-local means operators,and achieve adaptive regularization in our algorithm.
A PPENDIX
We introduce the process of how to solve the NLSU model in detail in this Appendix,which is an extension of Section III-B.Since the NLSU algorithm is based on the split augmented Lagrangian method of multipliers,and the ADMM strategy is adopted,we describe the process of the augmentation ?rst,and we then describe the splitting.
We ?rst use
,,,,,,and to simplify the representation of the original part in (12),e.g.,
,denotes .As the ADMM strategy is used to decompose the dif ?cult problem of (12)into a sequence of simpler ones [10],[34],the objective function can be written as (20):
(20)
The expanded augmented Lagrangian formula is then ob-tained in (21)to add the many ancillary terms by utilizing
the ADMM in the optimization problem of (20).For ex-ample,
as an ancillary term is related to ,is related to ,and so on.
is a constant denoting the Lagrangian multiplier in (21),and ,,,,,are sequences satisfying
[35]
(21)
The augmentation of the NLSU model has now been accom-plished.
In the following part,we elaborate on how to split the compli-cated optimization relation (21)based on the ADMM strategy,and give details of the derivation of the NLSU.
1904IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING,VOL.7,NO.6,JUNE
2014
Fig.14.Nuance data.(a)Nuance data.(b)Reference data.(c)The ROIs.(d)The classi ?cation result using SVM.(e)Spectral library for the Nuance
data.
Fig.15.Approximate true abundance images.(a)Dead leaves.(b)Fresh grass.(c)Background.
TABLE III P ARAMETER
S ETTINGS W ITH
THE
R EAL D
ATA
To achieve the solution of the augmented Lagrangian for-mula by performing an alternating minimization process,the partial derivative of (21)can be written as
(22)
where
(23)
and
is the identity matrix.
ZHONG et al.:NON-LOCAL SPARSE UNMIXING FOR HYPERSPECTRAL REMOTE SENSING IMAGERY 1905
TABLE IV
C OMPUTATIONAL T IMES W ITH THE
D IFFERENT M ETHODS FOR TH
E C UPRITE D
ATA
Fig.16.Estimated abundance fractions with the different methods for the Cuprite data.
The optimization of ,,,,,can be com-puted as follows,which is obtained by means of derivation.
First,we compute the optimization of
as follows:(24)
As the norm is not differentiable,in the case of sparse approximation of ,the solution is obtained by a soft shrinkage operator [36],as follows:
(25)
For
and ,the extended non-local split augmented La-grangian computation uses the non-local TV norm instead of the standard TV norm:
(26)
1906IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING,VOL.7,NO.6,JUNE2014
Fig.17.Estimated abundance fractions for each component with the different methods for the Nuance data.
TABLE V
SRE V ALUES A CHIEVED W ITH THE D IFFERENT M ETHODS FOR THE N UANCE D ATA
The solution of(26)(,,,and)is obtained by
performing an alternating minimization process:
(27)
The Euler-Lagrange equation for is given by
(28)
which provides
(29)Fig.18.SRE in relation to and.
ZHONG et al.:NON-LOCAL SPARSE UNMIXING FOR HYPERSPECTRAL REMOTE SENSING IMAGERY
1907
Fig.19.SRE in relation to (a)and (b)
.
Fig.20.Results obtained by SUnSAL,SUnSAL-TV ,NLSU (without ASC),
and NLSU (with ASC).
The that appears in is de ?ned as the vector of all the partial derivatives at ,which obtains the
non-local gradient [35],and
represents the graph Laplacian,which is negative semi-de ?nite.For more information about computing this,please refer to [35].
Similar to
,the optimization of can be obtained by applying the shrinkage operator [36]:
(30)
Finally,
and are acquired by minimizing the following
two problems:
(31)
According to the de ?nition of
and in (14)and (15),
the optimization of these two values can (32)
is a Kronecker product,which is de ?ned as
...
...
...
and ,.
Finally,the Lagrangian multipliers are updated:
(33)
All the variables are ?rst initialized by using:
where is the observed data.
A CKNOWLEDGMENT
The authors would like to thank the research group super-vised by Prof.J.M.Bioucas-Dias and Prof.A.Plaza for making the source code of the latest sparse algorithms,SUnSAL and
1908IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING,VOL.7,NO.6,JUNE2014
SUnSAL-TV,available to the community,and for the free public downloads of the A VIRIS image data(http://www.lx.it.pt/~bi-oucas/publications.html).The authors would also like to thank the editor,associate editor,and anonymous reviewers for their helpful comments,and Dr.M.D.Iordache for the helpful dis-cussions and constructive suggestions to improve this paper.
R EFERENCES
[1]N.Keshava and J.F.Mustard,“Spectral unmixing,”IEEE Signal Pro-
cessing Mag.,pp.44–57,Jan.2002.
[2]M.D.Iordache,J.M.Bioucas-Dias,and A.Plaza,“Sparse unmixing
of hyperspectral data,”IEEE Trans.Geosci.Remote Sens.,vol.49,no.
6,pp.2014–2039,2011.
[3]J.W.Boardman,F.A.Kruse,and R.O.Green,“Mapping target sig-
natures via partial unmixing of A VIRIS data,”in Proc.JPL Airborne
Earth Sci.Workshop,1995,pp.23–26.
[4]M.E.Winter,“N-FINDR:An algorithm for fast autonomous spec-
tral end-member determination in hyperspectral data,”in Proc.SPIE
Imaging Spectrometry V,2003,vol.3753,pp.266–275.
[5]H.Ren and C.-I.Chang,“Automatic spectral target recognition in hy-
perspectral imagery,”IEEE Trans.Aerosp.Electron.Syst.,vol.39,no.
4,pp.1232–1249,Oct.2003.
[6]H.Li and L.Zhang,“A hybrid automatic endmember extraction algo-
rithm based on a local window,”IEEE Trans.Geosci.Remote Sens.,
vol.49,no.11,pp.4223–4238,Nov.2011.
[7]R.A.Neville,K.Staenz,T.Szeredi,J.Lefebvre,and P.Hauff,“Au-
tomatic endmember extraction from hyperspectral data for mineral ex-
ploration,”in Proc.21st Can.Symp.Remote Sens.,1999,pp.21–24.
[8]M.Berman,H.Kiiveri,https://www.doczj.com/doc/756628821.html,gerstrom,A.Ernst,R.Dunne,and J.F.
Huntington,“ICE:A statistical approach to identifying endmembers in
hyperspectral images,”IEEE Trans.Geosci.Remote Sens.,vol.42,no.
10,pp.2085–2095,Oct.2004.
[9]J.M.Bioucas-Dias,A.Plaza,N.Dobigeon,M.Parente,D.Qian,P.
Gader,and J.Chanussot,“Hyperspectral unmixing overview:Geomet-
rical,statistical,and sparse regression-based approaches,”IEEE J.Sel.
Topics Appl.Earth Observ.Remote Sens.(JSTARS),vol.5,no.2,Apr.
2012.
[10]J.Bioucas-Dias and M.Figueiredo,“Alternating direction algorithms
for constrained sparse regression:Application to hyperspectral un-
mixing,”in Proc.2nd IEEE GRSS Workshop on Hyperspectral Image
and Signal Processing:Evolution in Remote Sensing(WHISPERS),
Reykjavik,Iceland,2010.
[11]M.D.Iordache,J.M.Bioucas-Dias,and A.Plaza,“Total variation
spatial regularization for sparse hyperspectral unmixing,”IEEE Trans.
Geosci.Remote Sens.,vol.50,no.11,pp.4484–4502,Nov.2012.
[12]A.Buades,B.Coll,and J.M.Morel,“A review of image denoising
algorithms,with a new one,”Multiscale Model.Sim.(SIAM Interdis-
ciplinary J.),vol.4,no.2,pp.490–530,Jul.2005.
[13]A.Buades,B.Coll,and J.M.Morel,“A non-local algorithm for image
denoising,”in IEEE Computer Society https://www.doczj.com/doc/756628821.html,puter Vision and Pat-
tern Recognition,2005,CVPR,2005,vol.2,pp.60–65.
[14]Y.Qian,M.Ye,and Q.Wang,“Noise reduction of hyperspectral im-
agery using nonlocal sparse representation with spectral-spatial struc-
ture,”in Proc.IEEE Int.Geosci.and Remote Sensing Symp.,IGARSS,
2012,pp.3467–3470.
[15]Y.Qian,Y.Shen,M.Ye,and Q.Wang,“3-D nonlocal means?lter with
noise estimation for hyperspectral imagery denoising,”in Proc.IEEE
Int.Geosci.and Remote Sensing Symp.,IGARSS,2012,pp.1345–1348.
[16]G.Gilboa and S.J.Osher,“Nonlocal linear image regularization and
supervised segmentation,”Multiscale Model.Sim.(SIAM Interdiscipli-
nary J.),vol.6,no.2,pp.595–630,Jul.2007.
[17]M.Protter,M.Elad,and H.Takeda,“Generalizing the non-local-means
to super-resolution reconstruction,”IEEE Trans.Image Process.,vol.
18,no.1,pp.36–51,Jan.2009.
[18]Y.Qian and M.Ye,“Hyperspectral imagery restoration using nonlocal
spectral-spatial structured sparse representation with noise estimation,”
IEEE J.Sel.Topics Appl.Earth Observ.Remote Sens.(JSTARS),vol.
6,no.2,pp.499–515,Apr.2013.
[19]J.Li et al.,“Classi?cation for hyperspectral imagery based on non-local
weighted joint sparsity model,”in Proc.4th Workshop Hyperspectral
Image Signal Processing:Evolution in Remote Sensing,2012,vol.1,
pp.1–4.
[20]M.Elad,Sparse and Redundant Representations:From Theory to Ap-
plications in Signal and Image Processing.New York,NY,USA:
Springer-Verlag,2010.
[21]A.M.Bruckstein,M.Elad,and M.Zibulevsky,“On the uniqueness
of non-negative sparse solutions to underdetermined systems of equa-
tions,”IEEE Trans.Inf.Theory,vol.54,no.11,pp.4813–4820,Nov.
2008.
[22]A.M.Bruckstein,D.L.Donoho,and M.Elad,“From sparse solutions
of systems of equations to sparse modeling of signals and images,”
SIAM Rev.,vol.51,no.1,pp.34–81,Feb.2009.
[23]E.J.Candès and T.Tao,“Decoding by linear programming,”IEEE
Trans.Inf.Theory,vol.51,no.12,pp.4203–4215,Dec.2005.
[24]E.J.Candès,J.Romberg,and T.Tao,“Stable signal recovery from in-
complete and inaccurate measurements,”Commun.Pure Appl.Math.,
vol.59,no.8,pp.1207–1223,Aug.2006.
[25]E.Candès and T.Tao,“Decoding by linear programming,”IEEE Trans.
Inf.Theory,vol.51,no.12,pp.4203–4215,Dec.2005.
[26]E.Candès and T.Tao,“Near-optimal signal recovery from random pro-
jections:Universal encoding strategies,”IEEE Trans.Inf.Theory,vol.
52,no.22,pp.5406–5424,Dec.2006.
[27]Y.C.Pati,R.Rezahfar,and P.S.Krishnaprasad,“Orthogonal matching
pursuit:Recursive function approximation with applications to wavelet
decomposition,”in Proc.27th Annu.Asilomar Conf.Signals Systems
and Computers,1993,vol.1,pp.40–44.
[28]S.Chen,D.Donoho,and M.Saunders,“Atomic decomposition by
basis pursuit,”SIAM Rev.,vol.43,no.1,pp.129–159,Feb.2001.
[29]D.Rogge,B.Rivard,J.Zhang,and J.Feng,“Iterative spectral un-
mixing for optimizing per-pixel endmember sets,”IEEE Trans.Geosci.
Remote Sens.,vol.44,no.12,pp.3725–3736,Dec.2006.
[30]G.Gilboa and S.Osher,“Nonlocal operators with applications to image
processing,”Multiscale Model.Simul.,vol.7,no.3,pp.1005–1028,
Nov.2008.
[31]M.D.Iordache,“A sparse regression approach to hyperspectral un-
mixing,”Ph.D.dissertation,https://www.doczj.com/doc/756628821.html,put.Eng.,IST,2011.
[32]J.Nascimento and J.M.Bioucas-Dias,“Vertex component analysis:
A fast algorithm to unmix hyperspectral data,”IEEE Trans.Geosci.
Remote Sens.,vol.43,no.4,pp.898–910,Apr.2005.
[33]X.Xu,Y.Zhong,L.Zhang,and H.Zhang,“Sub-pixel mapping based
on a MAP model with multiple shifted hyperspectral imagery,”IEEE
J.Sel.Topics Appl.Earth Observ.Remote Sens.(JSTARS),vol.6,no.
2,pp.580–593,Apr.2013.
[34]J.Eckstein and D.Bertsekas,“On the Douglas-Rachford splitting
method and the proximal point algorithm for maximal monotone
operators,”Math.Progr.,vol.5,pp.293–318,Apr.1992.
[35]X.Zhang,M.Burger,X.Bresson,and S.Osher,“Bregmanized
nonlocal regularization for deconvolution and sparse reconstruction,”
SIAM J.Imaging Sci.,vol.3,no.3,pp.253–276.
[36]J.F.Cai,S.Osher,and Z.Shen,“Convergence of the Linearized
Bregman Iteration for L1-Norm Minimization,”UCLA,2008,CAM
Report
08-52.
Yanfei Zhong(M’11)received the B.S.degree in in-
formation engineering and the Ph.D.degree in pho-
togrammetry and remote sensing from Wuhan Uni-
versity,China,in2002and2007,respectively.
He has been with the State Key Laboratory of
Information Engineering in Surveying,Mapping
and Remote Sensing,Wuhan University,since2007
and is currently a Professor.His research interests
include multi-and hyperspectral remote sensing
image processing,arti?cial intelligence,and pattern
recognition.He has published more than50research papers,including20peer-reviewed articles in international journals such as IEEE T RANSACTIONS ON G EOSCIENCE AND R EMOTE S ENSING and IEEE T RANSACTIONS ON S YSTEMS,M AN AND C YBERNETICS B.
Dr.Zhong was the recipient of the National Excellent Doctoral Dissertation Award of China(2009)and New Century Excellent Talents in University of China(2009).He was a Referee of IEEE T RANSACTIONS ON S YSTEMS,M AN AND C YBERNETICS B,IEEE J OURNAL OF S ELECTED T OPICS IN A PPLIED E ARTH O BSERV ATIONS AND R EMOTE S ENSING,and Pattern Recognition.
台式机没有网络连接如何解决 台式机没有网络连接解决方法一： 路由器通过网线连接到台式电脑上，台式电脑进入无线路由器管理者后台设置相应的上网参数，台式电脑需要有无线网卡，否则无法使用无线网络。 操作步骤参考如下： 接线方式：外网网线接modem的adsl网口，modem的lan网口分线到无线路由器的wan网口，无线路由器的lan网口分线到电脑网口; 电脑打开任意浏览器，输入 192.168.1.1(多数默认这个，具体详见路由器背面参数)，进入无线路由器后台登陆界面，输入相应的账号或是登陆密码登录即可(初次使用，账号或是登陆密码为默认，路由器背面可见。只需要登陆密码的，首次登陆自行设置即可); 进入无线路由器管理后台后，右侧界面一般会自动弹出运行向导界面，点击下一步(如未弹出，可点击左侧的运行向导); 上网方式设置，一般默认让路由器选择上网方式，点击下一步; 输入报装网线时，从运营商处得来的账号和密码，而不是无线路由器的账号和密码，输入后，点击下一步; 设置无线网络(wlan)的账号(ssid)和选择加密方式后设置密
码，点击下一步; 完成所有设置后，点击退出设置向导。wlan已经完成设置，可以正常使用。 台式电脑右键桌面右下角的网络和共享中心—本地连接—更改适配器—启用无线网络连接，将本地连接禁用即可。 备注：无线连接之后若出现黄色感叹号，将电脑无线网络的ip地址设置为自动获取即可。 台式机没有网络连接解决方法二： 第一步：确认无线网卡硬件已经安装好。 如果是台式机，一般是usb网卡，插在usb接口上即可。 如果是笔记本，无线网卡是自带的，一般不会接触不良。 第二步：确认无线网卡的驱动程序已经正确安装。 检查方法是：在桌面上的我的电脑上点右键，管理。 再点“设备管理器”，展开网络适配器前的+号。 如果电脑有一块有线网卡和一块无线网卡， 那么网络适配器下应该有2行， 如果是下图这样，就是无线网卡接触不良，或者没有安装无线网卡。 接触不良的问题，可以拆下无线网卡，清理干净后重新装回去。 还有一种情况，设备管理器中出现了问号或感叹号， 如果感叹号正好在无线网卡上，就说明无线网卡接触不良或者驱动程序没装好。 驱动程序的问题可以用驱动精灵或者鲁大师解决。
网卡灯不亮的原因及其解决方法 电脑不能上网怎么办？电脑网卡灯不亮了该如何解决呢？有用户在使用电脑时突然遇到断网问题，并且发现网口的灯不亮了，这是怎么回事？本文就给大家介绍电脑不能上网，网卡灯不亮的原因及其解决方法。 一、线路问题 首先分析下电脑网线是否损坏，网线插没插好，有没有接触不良的情况，是否有其他电器在使用干扰了网卡信号源的情况。本地电缆没有插好。电话线入户后连接接线盒，然后再到话音分离器分离，一线走电话、一线走MODEM（分离器上有标注）。同时确保线路通讯质量良好没有被干扰，没有连接其它会造成线路干扰的设备。并检查接线盒和水晶头有没有接触不良以及是否与其它电线串绕在一起。有条件最好用标准电话线，PC接ADSL Modem附带的双绞线。线路是影像上网的质量的重要因素之一。距离用户电缆线100米以内的无线电发射塔、电焊机、电车或高压电力变压器等信号干扰源，都能使用户下线接收杂波（铜包钢线屏蔽弱，接收信号能力强），对用户线引起强干扰。受干扰的信号往往是无屏蔽的下线部分进入，因为中继电缆有屏蔽层，干扰和影响都很小。如果在干扰大的地方用一些带屏蔽的下线，就会减少因干扰造成的速度不稳定或掉线现象。另外，电源线不可与adsl线路并行，以防发生串扰，导致adsl故障。另外其他也有很多因素造成网络不稳定，例如信号干扰、软件冲突。手机这一类辐射大的东西一定不要放在ADSL Modem的旁边，因为每隔几分钟手机会自动查找网络，这时强大的电磁波干扰足以造成ADSL Modem断流。 二、网卡问题 网卡是不是老化了，网使用双网卡有没有被冲突，电脑网卡灯不亮解决办法建议直接将网卡拔下来清洁下，清除主板上的插槽，重装网卡驱动或者更换网卡试下能不能恢复正常。 三、ADSL MODEM或者网卡设置问题 现在MODEM一般具有2种工作模式，一种是使用拨号软件的正常模式，一种是自动拨号的路由模式。在正常模式工作下，不需要对MODEM进行设置，使用默认即可。而路由模式则需要进行设置，MODEM 带有自己的闪存，可以将帐户、密码盒设置存入，进行开机自动拨号。此方法最常见的是设置错了ADSL Modem的IP地址，或是错误设置了DNS服务器。因为对于ADSL虚拟拨号的用户来说，是不需要设定IP地址的，自动分配即可。TCP/IP网关一般也不需要设置。但是设定DNS一定要设置正确，DNS 地址可以从当地电信部门获得。另外，TCP/IP设置最容易引起不能浏览网页的情况，一般设置为自动获得IP地址，但是DNS一定要填写。其他采用默认即可。 检测电话线是否有问题或者接触不良，质量差的电话线很容易造成掉线的。或者ADSL损坏啦，尝试不使用分离器而直接将外线接入ADSL Modem。分离器与ADSL Modem的连线不应该过长，否则不能同步。排除上述情况，只要重起ADSL Modem就可以解决同步问题。 四、系统出问题 也可能是系统出问题了。因为有的操作系统可能对ADSL的相关组件存在兼容性问题，这样可以到微软对系统进行升级，或者修复系统。有条件可以进行重装。如果软件有冲突就尽量找出冲突软件，对其卸载或者其他方法解决。 五、病毒破坏、防火墙、浏览器设置不正确
宽带连接错误的处理办法691、623、678、645、720、721、718、734、769、619、676、815(留着吧，不用麻烦别人喽） 宽带连接错误的处理办法 691、623、678、645、720、721、718、734、769、619、676、815(留着) 宽带连接错误691(由于域上的用户名或密码无效而拒绝访问)/错误635(未知错误)的处理 流程如下： （1）用户名密码填写错误 （2）如果用户帐号密码填写无误，则进入下一步继续处理；建议重建拨号软件（如果不懂装拨号软件，可参考户重新）： ?如果重装拨号软件后正常，原因为“拨号软件丢失”； ?如果重装拨号软件后故障依旧，请拨打客服电话10000电信(10060网通/10050铁通)进行障碍申告。 宽带连接错误６９１ 错误691真正意义上来讲：1：域上名出现错误，（用户名或密码输入错误）。2：服务器无反映，（机房用户端口错误，或帐号未被激活）。3：电话或宽带到期欠费造成。 出现错误691的原因 1.电信限制了你帐户使用数目，比如你这个帐户可以4个人用，现在4个人在用，你拨号就是错误691 2.你在用完后没断开，至少服务器那边还是没断开，以为你还在用，和上面情况类似，所以错误691 建议：每次关机的时候在宽带连接上右键，点断开 出现错误691后不要一直死缠烂打拨号，等待个几分钟再试试，如果一直出现这种情况，拨打客服电话10000电信(10060网通/10050铁通)，告诉工作人员你的电脑错误691上不了，然后她问“请问你的宽带编号多少”，告诉她你帐户，然后她有时问你开户的是谁，回答开户的是谁，然后过个几分钟她就会找人帮你搞定 宽带连接错误691的解决办法之一 解决ADSL莫名其妙的错误691问题 工厂的办公室里有无线路由器，而家里新装了宽带，还没有买路由器，所以只跟哥共用一个ADSL帐号上网，一根网线在两台手提之间插来拔去的。但是让人郁闷的是，插在哥
台式机没有网卡怎么办 台式机没有网卡解决方法一： 1、你的电脑没网卡驱动，就连不上网络，所以要找其他可以联网的电脑，下载你的网卡驱动，然后拷贝过来装上就行了。 2、在桌面右键单击我的电脑，选择管理，在弹出的框中左边栏选择设备管理器单击，再在右边框中选择网络适配器单击，就能够看见你的网卡的型号了。 3、到相应的官方网站下载该网卡的驱动程序来安装。集成网卡的话看主板型号。然后把型号记下来，去有网的地方下载。下载后拷贝到u盘，再插入电脑安装。最好到主板的官网去下载驱动。非集成网卡，直接看网卡品牌，型号下载方法同上。 台式机没有网卡解决方法二： 因为没有无线网卡和没有无线网卡驱动，所以就没有无线网络连接。要使台式电脑能够接收无线信号，必须安装无线网卡。 无线网卡按照接口的不同可以分为多种： 一种是台式机专用的pci接口无线网卡。 一种是usb无线网卡，这种网卡不管是台式机用户还是笔记本用户，只要安装了驱动程序，都可以使用。在选择时要注意的只有采用usb2.0接口的无线网卡才能满足802.11g或802.11g+的需求。 台式机没有网卡解决方法三：
网卡驱动可能需要重装，但usb驱动一般系统都自带。 如果手头有其他网卡，换一张网卡，使用系统直接支持的网卡。 网卡一般配有光盘驱动，使用原光盘安装驱动。 如果有刻录机，下载对应的网卡和usb驱动，刻录到光盘，再安装。 如果不介意重装一遍系统，可安装win7、win8等系统，他们本身自带的驱动包比较全面，安装后应该会自动安装网卡和usb 驱动。 相关阅读： 网卡一般设置 网卡属性设置步骤如下： 1)将"本地连接 2"改名为"控制网 a"，用于连接过程控制网 a 网，其属性设置如下： ip 地址：128.128.1.x(x 为操作节点地址限定范围内的值)，其它如 dns、wins 等设置为默认。 2)将"本地连接 3"改名为"控制网 b"，用于连接过程控制网 b 网，其属性设置如下： ip 地址：128.128.2.x(x 为操作节点地址限定范围内的值)，其它同上。 3)将"本地连接"改名为"操作网"，用于连接操作网，其属性设置如下： ip 地址：128.128.5.x(x 为操作节点地址限定范围内的值)，其它同上。
电脑网络不通是令我们头疼的一个问题，特别是小企业或者公司网络，偶尔要传一下文件什么的，如果出现网络不同或者不能上网等问题，我们可能就因为网络不通耽误了我们的大事。接下来我们来介绍一下，网络不通的几种可能和解决的办法！ 一、连接指示灯不亮 观察网卡后侧RJ45一边有两个指示灯。它们分别为连接状态指示灯和信号传输指示灯，其中正常状态下连接状态指示灯呈绿色并且长亮，信号指示灯呈红色，正常应该不停的闪烁。如果我们发现连接指示灯，也就是绿灯不亮，那么表示网卡连接到HUB或交换机之间的连接有故障。对此可以使用测试仪进行分段排除，如果从交换机到网卡之间是通过多个模块互连的，那么可以使用二分法进行快速定位。而一般情况下这种故障发生多半是网线没有接牢、使用了劣质水晶头等原因。而且故障点大多是连接的两端有问题，例如交换机的端口处和连接计算机的网卡处的接头，借助测试仪可以很轻松的就以找出故障进行解决。 二、信号指示灯不亮 如果信号指示灯不亮，那么则说明没有信号进行传输，但可以肯定的是线路之间是正常的。那么不防使用替换法将连接计算机的网线换到另外一台计算机上试试，或者使用测试仪检查是否有信号传送，如果有信号传送那么则是本地网卡的问题。在实际的工作经验证明网卡导致没有信息传送是比较普遍的故障。对此可以首先检查一下网卡安装是否正常、IP设置是否错误，可以尝试Ping一下本机的IP地址，如果能够Ping通则说明网卡没有太大问题。如果不通，则可以尝试重新安装网卡驱动来解决，另外对于一些使用了集成网卡或质量不高的网卡，容易出现不稳定的现象，即所有设置都正确，但网络却不通。对此可以将网卡禁用，然后再重新启用的方法，也会起到意想不到的效果。 另外，提醒各位想要网络稳定还是建议安装电信宽带，在广东电信网厅参加双11活动，宽带打5折，还送路由器，不用安装费，续约打个95折。 三、降速使用 很多网卡都是使用10M/100M自适应网卡，虽然网卡的默认设
电脑没有网卡解决办法 一、检查网络线路连接和网卡是否良好。 二、安装网卡驱动。 1、右击“我的电脑”----“属性”---“硬件”----“设备管理器”—展开“网络适配器”—看有没有黄色的问号？，有，说明缺网卡驱动，有“！”号，说明该驱动不能正常使用，将其卸载。（注意要记下，这是你使用的网卡型号）。 2、将网卡驱动光盘放入光驱，右击“我的电脑”--“属性”--“硬件”--“设备管理器”，展开“网络适配器”，右击网卡—选“更新驱动程序”，打开“硬件更新向导”，选“是，仅这一次”--“下一步”--“自动安装软件”--“下一步”，系统即自动搜索并安装光盘中的网卡驱动程序，如果该光盘没有适合你用的网卡驱动，再换一张试试，直到完成。 3、如果没有适合的光盘，到朋友家，从驱动之家、中关村在线、华军等网站下载驱动软件，下载驱动软件要注意：一是品牌型号要对，二是在什么系统上便用，三是要看该驱动软件公布的时间，最新的未必适合使用，可多下载几个，拷到U盘上，挑着使。 4、将U盘插在自家的上，下载的驱动软件一般有自动安装功能，打开即自动安装。不能自动安装的，解压后备用，要记下该软件在磁盘中的具体路径，如D：\ ……\……。右击“我的电脑”--“属性”--“硬件”--“设备管理器”，展
开“网络适配器”右击网卡，选“更新驱动程序”，打开“硬件更新向导”，去掉“搜索可移动媒体”前的勾，勾选“从列表或指定位置安装”---“下一步”，勾选“在搜索中包括这个位置”，在下拉开列表框中填写要使用的声卡驱动文件夹的路径（D：\……\……--“下一步”，系统即自动搜索并安装你指定位置中的网卡驱动程序。 三、拨号上网，还不行，系统文件丢失，修复或重装系统。 四、可能是这个网卡坏了，换一个网卡装上驱动，再试上网。
电脑网络不通是令我们头疼的一个问题，特别是小企业或者公司网络，偶尔要传一下文件什么的，如果出现网络不同或者不能上网等问题，我们可能就因为网络不通耽误了我们的大事。接下来我们来介绍一下，网络不通的几种可能和解决的办法！ 一、连接指示灯不亮 观察网卡后侧RJ45一边有两个指示灯。它们分别为连接状态指示灯和信号传输指示灯，其中正常状态下连接状态指示灯呈绿色并且长亮，信号指示灯呈红色，正常应该不停的闪烁。如果我们发现连接指示灯，也就是绿灯不亮，那么表示网卡连接到HUB或交换机之间的连接有故障。对此可以使用测试仪进行分段排除，如果从交换机到网卡之间是通过多个模块互连的，那么可以使用二分法进行快速定位。而一般情况下这种故障发生多半是网线没有接牢、使用了劣质水晶头等原因。而且故障点大多是连接的两端有问题，例如交换机的端口处和连接计算机的网卡处的接头，借助测试仪可以很轻松的就以找出故障进行解决。 提醒各位，想要网速快、稳定，推荐到广东电信网厅参加双十一活动，装宽带比较便宜，不用安装费，送路由器，续费的话95折。 二、信号指示灯不亮 如果信号指示灯不亮，那么则说明没有信号进行传输，但可以肯定的是线路之间是正常的。那么不防使用替换法将连接计算机的网线换到另外一台计算机上试试，或者使用测试仪检查是否有信号传送，如果有信号传送那么则是本地网卡的问题。在实际的工作经验证明网卡导致没有信息传送是比较普遍的故障。对此可以首先检查一下网卡安装是否正常、IP设置是否错误，可以尝试Ping一下本机的IP地址，如果能够Ping通则说明网卡没有太大问
题。如果不通，则可以尝试重新安装网卡驱动来解决，另外对于一些使用了集成网卡或质量不高的网卡，容易出现不稳定的现象，即所有设置都正确，但网络却不通。对此可以将网卡禁用，然后再重新启用的方法，也会起到意想不到的效果。 三、降速使用 很多网卡都是使用10M/100M自适应网卡，虽然网卡的默认设置是“自适应”，但是受交换机速度或网线的制作方法影响，可能出现一些不匹配的情况。这个时候不防试试把网卡速度直接设为10M。其方法是右击“本地连接”打开其属性窗口，在“常规”选项卡中单击“配置”按钮，将打开的网卡属性窗口切换到“高级”选项卡，在“属性”列表中选中“Link Speed/Duplex Mode”，在右侧的“值”下拉菜单中选择“10 Full Mode”，依次单击“确定”按钮保存设置。 四、防火墙导致网络不通 在局域网中为了保障安全，很多朋友都安装了一些防火墙。这样很容易造成一些“假”故障，例如Ping不通但是却可以访问对方的计算机，不能够上网却可以使用QQ等。判断是否是防火墙导致的故障很简单，你只需要将防火墙暂时关闭，然后再检查故障是否存在。而出现这种故障的原因也很简单，例如用户初次使用IE访问某个网站时，防火墙会询问是否允许该程序访问网络，一些用户因为不小心点了不允许这样以后都会延用这样的设置，自然导致网络不通了。比较彻底的解决办法是在防火墙中去除这个限制。例如笔者使用的是金山网镖，那么则可以打开其窗口，切换到“应用规则”标签，然后在其中找到关于
无线网卡搜不到信号的解决方法 今天，心血来潮，想搜一下有没有没加密的无线信号，也蹭个“网” 于是打开无线网络连接，一搜，没有信号！！ 打开同学的电脑一搜，有信号（不过加密了！hfwang1253）。我很郁闷~~~ 虽然上不了网，但自己的无线却搜索不到信号？？问题很严重，必需解决！！！ 于是开始查看问题是出在哪里了。经过不懈努力，终于搞定了，拿出来的与大家分享一下下希望可以解决大家的问题，如有不异意，请提出来！！谢谢。 一、首先，我们需要一个配置好的无线路由器并且开启了无线功能。 出于安全考虑我们还可以给无线路由器发出的信号进行加密，防止别人“蹭网” 设置好后可以看路由器上面的无线信号指示灯是否是亮的（一般都有这个指示灯） 如果是亮的或者是一闪一闪的一般都不会有问题了。 二、开启电脑的无线功能。 1.我们需要有一台装有无线网卡的电脑，并且装好驱动（硬件肯定不能有问题的） 笔记本一般都带有无线网卡；如果是台式机，可以通过安装一个USB口的无线信号接收装置（一般买无线都会送一个） 2.如果做好了以上操作，那么我们可以打开“网上邻居”，然后点击“查看网络连接”来查看网络连接情况 如果正常的安装了网卡，就会在这里出现网络连接的图标。 如果没有则需要查看网卡驱动是否安装正确 注：如果点击“查看网络连接”弹出如下提示 说明我们禁用了“ Network Connections ”此服务，只需启用此服务就要解决此问题 开启“Network Connections”服务： 1》开始----运行----输入:services.msc----确定----打开“Network Connections” 2》打到并双击“Network Connections”服务，弹出对话框 3》将“启动类型”设置为“自动” 4》然后点击“启动”，待“服务状态”显示为“启用”时点“确定” 如：
win7电脑网卡消失不见怎么办 一、网吧找不到网络适配器原因分析之irq中断引起故障 现在pci网卡均支持即插即用，在安装驱动时会自动分配irq(中断)资源。如果预定的irq资源被声卡、modem、显卡等设备占用，而系统又不能给网卡重新指定另外的irq资源的话，就会发生设备冲突，导致设备不能使用的问题。如realtekrt8029pciethernet网卡就容易和显示卡发生冲突(均使用irq10)。解决方法很简单，我们可以查找一下主板说明书中对pci插槽优先级部分的说明,将冲突的设备更换到优先级更高的pci插槽上(一般来说，越靠近agp插槽的pci插槽，优先级别就越高)，并进行调换，直到两冲突设备不再冲突为止。这种方法很简单，但相对来说就会比较繁复。除此之外，我们还可以在网卡的设备属性里面，手动为网卡重新分配irq值： 第一步：按ctrl+pause快捷键打开“系统属性”，再依次单击“硬件→设备管理”打开设备管理器 第二步：在设备管理器中，展开“网络适配器”，双击网卡设备打开网卡的设备属性，并在“资源”选项卡的“资源设置”列表中选择“中断请求”，然后取消“使用自动的设置”复选框。 第三步：单击“更改设置”按钮，重新为网卡分配一个irq 值，直到“冲突设备列表”显示为“没有冲突”即可。 二、更换位置造成网卡冲突现象
在安装有windows2000以上版本的计算机系统中，要是我们没有先对网卡进行卸载操作，而是直接挪动网卡的安装插槽位置，并将以前使用的ip地址重新分配给新位置处的网卡时，我们就会发现系统提示网卡ip地址发生了冲突现象，网卡冲突也会导致网吧找不到网络适配器。 要想消除网卡冲突故障现象，我们只有想办法将原来的网卡删除掉或者强行将以前网卡使用的ip地址释放出来，最简单的操作方法莫过于将网卡的安装位置重新返回到原来的主板插槽中，然后打开系统的设备管理器窗口，找到原来网卡设备对应的图标，然后用鼠标右键单击该图标，从其后出现的右键菜单中执行“卸载”命令。这样就能将旧网卡设备从计算机系统中先删除掉了。之后再将网卡设备插入到新的插槽位置，再为其分配以前使用的ip地址，这样就可以消除网卡冲突现象了。 5当然，在安装有windows2000系统的计算机中，我们也能够借助系统的“添加/删除硬件”功能，将以前的旧网卡设备先从计算机系统中删除掉。依次单击“开始”/“设置”/“控制面板”命令，在其后出现的控制面板窗口中双击“添加/删除硬件”图标，打开添加/删除硬件向导界面，找到并选中该界面中的“卸载/拔掉硬件——卸载设备”项目，同时选中向导界面中的“显示隐藏设备”项目，这样我们就能轻易在设备管理器中找到隐藏起来的旧网卡设备，然后按常规方法就能将旧网卡设备从计算机系统中删除掉了。 对于那些安装了windowsxp系统的计算机来说，由于系统自身没有内置“添加/删除硬件”功能，不过我们可以通过修改注册
网络连接不稳定 在网卡工作正常的情况下，网卡的指示灯是长亮的(而在传输数据时，会快速地闪烁)。如果出现时暗时明，且网络连接老是不通的情况，最可能的原因就是网卡和PCI插槽接触不良。和其他PCI设备一样，频繁拔插网卡或移动电脑时，就很容易造成此类故障，重新拔插一下网卡或换插到其他PCI插槽都可解决。此外，灰尘多、网卡“金手指”被严重氧化，网线接头(如水晶头损坏)也会造成此类故障。只要清理一下灰尘、用报纸把“金手指”擦亮即可解决。 驱动程序出现的故障 网卡和其他硬件一样，驱动程序不完善也极易引起故障，比如采用瑞昱(Realtek)RTL8469芯片的网卡，在Windows 2000下就经常会出现NetBIOS TCP/IP方面的错误。将驱动更新，此类问题就会迎刃而解。所以，当网卡出现一些不明缘由的故障时，可以到“驱动之家”(www．mydrivers．com)等专业网站更新驱动来解决(推荐大家优先使用经过微软WHQL 认证的驱动，通过此认证的驱动程序与Windows系统的兼容性是最好的)。一般在排除硬件、网络故障前提下，升级或重装驱动可以解决很多莫名故障。如果网卡故障是发生在驱动程序更新之后的话，可以用网卡自带的驱动程序来恢复一下。 IRQ中断引起故障 现在PCI网卡均支持即插即用，在安装驱动时会自动分配IRQ(中断)资源。如果预定的IRQ资源被声卡、Modem、显卡等设备占用，而系统又不能给网卡重新指定另外的IRQ资源的话，就会发生设备冲突，导致设备不能使用的问题。如Realtek RT8029 PCI Ethernet网卡就容易和显示卡发生冲突(均使用IRQ10)。解决方法很简单，我们可以查找一下主板说明书中对PCI插槽优先级部分的说明,将冲突的设备更换到优先级更高的PCI插槽上(一般来说，越靠近AGP插槽的PCI插槽，优先级别就越高)，并进行调换，直到两冲突设备不再冲突为止。这种方法很简单，但相对来说就会比较繁复。除此之外，我们还可以在网卡的设备属性里面，手动为网卡重新分配IRQ值： 第一步：按Ctrl+Pause快捷键打开“系统属性”，再依次单击“硬件→设备管理”打开设备管理器。 第二步：在设备管理器中，展开“网络适配器”，双击网卡设备打开网卡的设备属性，并在“资源”选项卡的“资源设置”列表中选择“中断请求”，然后取消“使用自动的设置”复选框。 第三步：单击“更改设置”按钮，重新为网卡分配一个IRQ值，直到“冲突设备列表”显示为“没有冲突”即可(如图1)。 网络连接不稳定 在网卡工作正常的情况下，网卡的指示灯是长亮的(而在传输数据时，会快速地闪烁)。如果出现时暗时明，且网络连接老是不通的情况，最可能的原因就是网卡和PCI插槽接触
解决电脑双网卡无法上网的问题 有的公司或机构的内网是封闭的，只有内部网络，没有出口网关，上不了外网的。但有时候我们希望再加一块网卡连上外网线，让电脑既能访问内网，又能访问外网（互联网）。然而这两块网卡好像不能共存，要上外网的时候只能禁用内网卡，要上内网时只能禁用外网卡，有没有办法实现内外网共存呢？ 可以实现的。 首先网络拓扑如下： 有双网卡的电脑原来只有一块网卡的时候，那块内网卡的IP是公司分配的。该电脑被划分在vlan10里，分配了一个IP：192.168.10.5/24位掩码，网关为192.168.10.1 后来想让这台电脑能访问互联网，于是又加了一块外网卡，接到能上互联网的路由器上，该路由器默认分配的IP是192.168.1.0/24网段的，例如外网卡获取到的IP是192.168.1.14，默认网关是192.168.1.1在有双网卡的电脑上的cmd命令行里输入ipconfig回车：
上图显示确实有两块网卡，且IP地址都是正确的，都有网关，ping各自的网关都能通 当两块网卡都启用时，却不能上网了（不能上网的情况有两种） 1）有时是不能访问内网的服务器（IP为192.168.30.254），但可以上外网 2）有时是可以访问内网服务器，而不能访问外网 因为这两块网卡都配有默认网关，当要访问的目的IP与这两块网卡的IP不在同一个网段时，计算机就会把数据发给默认网关，从网关出去了。有时候能ping通内网的服务器是因为该数据包从内网的网关出去了，而ping外网IP不通，是因为该数据包也从内网网关出去了，而内网又没有出口到互联网去，所以不通。 同理，有时能ping通外网而不通内网，是因为数据包都从外网网关出去了。 也就是说，当目的IP与本机的网卡IP在同一个网段时，都能ping通，而不在一个网段时，会从网关转发出去，那么到底从哪个网卡的网关出去呢？ 这得看计算机的路由表，在cmd命令行输入route-4print １）当内网卡是连网线，外网卡是一块无线网卡时，看到的路由表可能如下： 内网卡IP为192.168.10.5外网卡IP为192.168.1.14
Linux系统\Centos没有网卡eth0配置文件解决方法 1.临时解决办法 设置一个临时的IP地址 ifconfig eth0 192.168.100.1 设置完成后立即可生效，但不要重启，否则就不生效了 再使用ifconfig /ip 查看设置的IP地址即可。 2.永久解决方法 一般来说如果没有网卡eth0配置文件，你到网卡配置目/etc/sysconfig/network-scripts/下面都会有一个 ifcfg-lo 回环接口配置文件，拷贝一份 cd /etc/sysconfig/network-scripts/ cp ifcfg-lo ifcfg-etho 然后更改ifcfg-etho的配置文件即可 首先记得用ifconfig看一下Eth0的MAC地址 Viifcfg-eth0 DEVICE=eth0 ONBOOT=yes IPADDR=192.168.100.1 BOOTPROTO=static NETWORK=192.168.100.0 NETMASK=255.255.255.0 BROADCAST=192.168.100.255 DNS1=8.8.8.8
TYPE=Ethernet GATEWAY=192.168.100.254 IPV6INIT=no HWADDR=00:0C:29:66:BB:3E NAME="System eth0" 设置好网卡配置之后，需要把网络服务重启一下。输入如下命令: service network restart 如果没有提示任何错误则表示设置格式基本没有问题哈最后可以尝试一下使用reboot命令重启虚拟机后是否生效。
以下是导致网络缓慢的常见问题，以及一些常见网络问题的解决方法，在此整理给各位网友。 一、网络自身问题 您想要连接的目标网站所在的服务器带宽不足或负载过大。处理办法很简单，请换个时间段再上或者换个目标网站。 二、网线问题导致网速变慢 我们知道，双绞线是由四对线按严格的规定紧密地绞和在一起的，用来减少串扰和背景噪音的影响。同时，在T568A标准和T568B标准中仅使用了双绞线的1、2和3、6四条线，其中，1、2用于发送，3、6用于接收，而且1、2必须来自一个绕对，3、6必须来自一个绕对。只有这样，才能最大限度地避免串扰，保证数据传输。本人在实践中发现不按正确标准(T586A、T586B)制作的网线，存在很大的隐患。表现为：一种情况是刚开始使用时网速就很慢;另一种情况则是开始网速正常，但过了一段时间后，网速变慢。后一种情况在台式电脑上表现非常明显，但用笔记本电脑检查时网速却表现为正常。对于这一问题本人经多年实践发现，因不按正确标准制作的网线引起的网速变慢还同时与网卡的质量有关。一般台式计算机的网卡的性能不如笔记本电脑的，因此，在用交换法排除故障时，使用笔记
本电脑检测网速正常并不能排除网线不按标准制作这一问题的存在。我们现在要求一律按T586A、T586B标准来压制网线，在检测故障时不能一律用笔记本电脑来代替台式电脑。 三、网络中存在回路导致网速变慢 当网络涉及的节点数不是很多、结构不是很复杂时，这种现象一般很少发生。但在一些比较复杂的网络中，经常有多余的备用线路，如无意间连上时会构成回路。比如网线从网络中心接到计算机一室，再从计算机一室接到计算机二室。同时从网络中心又有一条备用线路直接连到计算机二室，若这几条线同时接通，则构成回路，数据包会不断发送和校验数据，从而影响整体网速。这种情况查找比较困难。为避免这种情况发生，要求我们在铺设网线时一定养成良好的习惯：网线打上明显的标签，有备用线路的地方要做好记载。当怀疑有此类故障发生时，一般采用分区分段逐步排除的方法。 四、网络设备硬件故障引起的广播风暴而导致网速变慢 作为发现未知设备的主要手段，广播在网络中起着非常重要的作用。然而，随着网络中计算机数量的增多，广播包的数量会急剧增加。当广播包的数量达到30%时，网络的传输效率将会明显下降。当网卡或网络设备损坏后，会不停地发送广播包，从而导致广播风暴，使网络通信陷于瘫痪。因此，当网络设备硬件有故障时也会引起网速变慢。当怀疑有此类故障时，首先可采用置换法替换集线器或交换机来排除集线设备故障。如果这些设备没有
现在电脑的首要功能就是上网，而生活和工作中经常碰到的上网问题却是很多的，下面就来说说电脑无法上网的解决办法。 1、网线问题：首先来看电脑桌面的右下角有一个小显示器的图标，这就是网络标志， 它如果显示一个红色的叉号，绝大部分就是网线没有插好，极个别的是网卡坏。先将连接电脑的网线两头重新插拔，注意检查网猫和路由器或者交换机是否工作正常；其次检查网线是否有断裂，或者水晶头损坏；最后可以更换一根网线进行测试；其实还可以使用测线仪测一下网线通不通，但是一般家庭没有这个设备，这里就不讲了；如果都不行，依然显示叉号，那就是网卡坏了，需要更换或维修，可以考虑USB接口的网卡，但是可能对网速稍有影响（很小），对网速要求高的就找专业的人进行维修吧。 2、网络设置问题：这个情况比较复杂，讲几个比较常见的吧；右下角网络图标显示一 个黄色的叹号，这证明网线是接通了的，但是无法正常连接网络，第一， 如果是通过路由器连接的网络，检查网猫是否工作正常，可以将猫和路由器断电重启测试，再看看外来的主线是否有信号具体可以找网络提供运营商咨询是否有网络故障；如果是办公环境，使用的交换机，可以将交换机断电重启，或者检查交换机到网络主设备的网线是否连接正常；第二，在桌面上的“网络”图标上点击鼠标右键再点属性，在弹出窗口左上方选择“更改适配器设置”，找到“本地连接”，点右键选禁用，然后再点右键选启用，进行尝试；
或者在网络属性窗口点击黄色叹号标志，系统自动检测进行修复一下； 第二，看一下IP地址，在本地连接图标上点右键，点属性，在对话框里找到“I nternet 协议版本4（TCP/I pv4）”双击，看一下IP地址是否正确，一般家庭网络都是在自动获取IP 地址和自动获取DNS服务器地址上；有的办公环境里可能会使用固定IP地址，这就需要咨询一下你们的网络管理人员，确认地址是否有错误，默认的DNS一般是有两种，8.8.8.8和114.114.114.114，或者根据办公使用的网络设备而定。 无法上网的问题暂时先写到这，后面想起来再补充。
安装网卡驱动以后无法上网的解决办法 宽带上网故障诊断方法如下（以ADSL为例） （1）首先检查电话线有无问题（可以拨一个电话测试），如果正常，接着检查信号分频器是否连接正常（其中电话线接Line口，电话机接Phone口，ADSL Modem 接Modem口）。 （2）如果信号分频器的连接正常，接着检查ADSL Modem的“Power（电源）”指示灯是否亮，如果不亮，检查`ADSL Modem电源开关是否打开，外置电源是否插接良好等。 （3）如果亮，接着检查“LINK（同步）”指示灯状态是否正常（常亮为闪烁）；如果不正常，检查ADSL Modem的各个连接线是否正常（从信号分频器连接到ADSL Modem的连线是否接在Line口，和网卡连接网线是否接在LAN口，是否插好），如果连接不正确，重新接好连接线。 （4）如果正常，接着检查“LAN”或“PC”指示灯状态是否正常。如果不正常，检查ADSL Modem上的LAN口是否接好，如果接好，接着测试网线是否正常，如果不正常，更换网线；如果正常，将电脑和ADSL Modem关闭30秒后，重新开启ADSL Modem和电脑电源。 （5）如果故障依旧，接着依次单击“开始／控制面板／系统／硬件／设备管理器”命令，打开“设备管理器”窗口中双击“网络适配器”选项，打开“网络适配器属性”对话框，然后检查网卡是否有冲突，是否启用，如果网卡有冲突，调整网卡的中断值。 （6）如果网卡没有冲突，接着检查网卡是否正常（是否接触不良、老化、损坏等），可以用替换法进行检测，如果不正常，维修或更换网卡。 （7）如果网卡正常，接着在“网上邻居”图标上单击鼠标右键，在打开的快捷菜单中选择“属性”命令，打开“本地连接属性”对话框。在“本地连接属性”对话框中检查是否有“Internet协议（TCP/IP）”选项。如果没有，则需要安装该协议；如果有，则双击此项，打开“Internet协议（TCP/IP）属性”对话框，然后查看IP地址、子网掩码、DNS的设置，一般均设为“自动获取”。 （8）如果网络协议正常，则为其他方面故障，接着检查网络连接设置、浏览器、PPPOE协议等方面存在的故障。
台式机没有网卡驱动怎么办 台式机没有网卡驱动解决方法一： 1、集成网卡的话看主板型号。 2、然后把型号记下来，去有网的地方下载。 3、下载后拷贝到u盘，再插入电脑安装。 4、最好到主板的官网去下载驱动。 5、非集成网卡，直接看网卡品牌，型号下载方法同上。 台式机没有网卡驱动解决方法二： 系统有自带的网卡驱动的，可以尝试一下以下操作 控制面板-添加硬件-下一步-是，已连接硬件-选择带问号的网卡，下一步-厂商选择microsoft，网卡选择对应你的网卡类型，不知道就一个个试。 台式机没有网卡驱动解决方法三： 1、你的电脑没网卡驱动，就连不上网络，所以要找其他可以联网的电脑，下载你的网卡驱动，然后拷贝过来装上就行了。 2、在桌面右键单击我的电脑，选择管理，在弹出的框中左边栏选择设备管理器单击，再在右边框中选择网络适配器单击，就能够看见你的网卡的型号了。 3、到相应的官方网站下载该网卡的驱动程序来安装。集成网卡的话看主板型号。然后把型号记下来，去有网的地方下载。下载后拷贝到u盘，再插入电脑安装。最好到主板的官网去下载驱
动。非集成网卡，直接看网卡品牌，型号下载方法同上。 相关阅读： 网卡功能详解 网卡上面装有处理器和存储器(包括ram和rom)。网卡和局域网之间的通信是通过电缆或双绞线以串行传输方式进行的。而网卡和计算机之间的通信则是通过计算机主板上的i/o总线以并行传输方式进行。 因此，网卡的一个重要功能就是要进行串行/并行转换。由于网络上的数据率和计算机总线上的数据率并不相同，因此在网卡中必须装有对数据进行缓存的存储芯片。 在安装网卡时必须将管理网卡的设备驱动程序安装在计算机的操作系统中。这个驱动程序以后就会告诉网卡，应当从存储器的什么位置上将局域网传送过来的数据块存储下来。网卡还要能够实现以太网协议。 网卡并不是独立的自治单元，因为网卡本身不带电源而是必须使用所插入的计算机的电源，并受该计算机的控制。因此网卡可看成为一个半自治的单元。当网卡收到一个有差错的帧时，它就将这个帧丢弃而不必通知它所插入的计算机。 当网卡收到一个正确的帧时，它就使用中断来通知该计算机并交付给协议栈中的网络层。当计算机要发送一个ip数据包时，它就由协议栈向下交给网卡组装成帧后发送到局域网。 随着集成度的不断提高，网卡上的芯片的个数不断的减少，虽然各个厂家生产的网卡种类繁多，但其功能大同小异看了“台式机没有网卡驱动怎么办”文章的
新装Linux系统没有网卡驱动的解决办法 你还在为不知道新装Linux系统没有网卡驱动而不知所措么?新装Linux系统没有网卡驱动你了解多少?下面来是小编为大家收集的新装Linux系统没有网卡驱动的解决办法，欢迎大家阅读： 新装Linux系统没有网卡驱动的解决办法 首先说明几个命令： #lsmod ——显示已载入系统的模块。 执行lsmod指令，会列出所有已载入系统的模块。Linux 操作系统的核心具有模块化的特性，应此在编译核心时，务须把全部的功能都放入核心。你可以将这些功能编译成一个个单独的模块，待需要时再分别载入。 #dmidecode
——以一种可读的方式dump出机器的DMI(Desktop Management Interface)信息。这些信息包括了硬件以及BIOS，既可以得到当前的配置，也可以得到系统支持的最大配置，比如说支持的最大内存数等。 #lspci ——list all PCI devices:列出机器中的PCI 设备(声卡、显卡、Modem、网卡、USB、主板集成设备也能列出来)，通过该命令可以查到网卡的厂商和型号。 #modprobe ——自动处理可载入模块。modprobe可载入指定的个别模块，或是载入一组相依的模块。modprobe会根据depmod所产生的相依关系，决定要载入哪些模块。若在载入过程中发生错误，在modprobe会卸载整组的模块。 #depmod ——分析可载入模块的相依性。depmod可检测模块的相依性，供modprobe在安装模块时使用。
八种常见网络不通的解决办法 搭建一个网络是相对比较轻松的事情，但是要想管理好网络却是一件比较繁琐的工作。且不说用户给你提出的各种要求会让你忙个不停，单单网络不通这一故障解决起来就够麻烦的，因为导致网络不通的原因实在太多。但是作为网络管理者也不必为此而失去解决问题的勇气，只要我们将故障进行现象进行规类，还是有规律可循的。 一、连接指示灯不亮 观察网卡后侧RJ45一边有两个指示灯。它们分别为连接状态指示灯和信号传输指示灯，其中正常状态下连接状态指示灯呈绿色并且长亮，信号指示灯呈红色，正常应该不停的闪烁。如果我们发现连接指示灯，也就是绿灯不亮，那么表示网卡连接到HUB或交换机之间的连接有故障。对此可以使用测试仪进行分段排除，如果从交换机到网卡之间是通过多个模块互连的，那么可以使用二分法进行快速定位。而一般情况下这种故障发生多半是网线没有接牢、使用了劣质水晶头等原因。而且故障点大多是连接的两端有问题，例如交换机的端口处和连接计算机的网卡处的接头，借助测试仪可以很轻松的就以找出故障进行解决。 二、信号指示灯不亮 如果信号指示灯不亮，那么则说明没有信号进行传输，但可以肯定的是线路之间是正常的。那么不防使用替换法将连接计算机的网线换到另外一台计算机上试试，或者使用测试仪检查是否有信号传送，如果有信号传送那么则是本地网卡的问题。在实际的工作经验证明网卡导致没有信息传送是比较普遍的故障。对此可以首先检查一下网卡安装是否正常、IP设置是否错误，可以尝试Ping一下本机的IP地址，如果能够Ping通则说明网卡没有太大问题。如果不通，则可以尝试重新安装网卡驱动来解决，另外对于一些使用了集成网卡或质量不高的网卡，容易出现不稳定的现象，即所有设置都正确，但网络却不通。对此可以将网卡禁用，然后再重新启用的方法，也会起到意想不到的效果。 三、降速使用 很多网卡都是使用10M/100M自适应网卡，虽然网卡的默认设置是“自适应”，但是受交换机速度或网线的制作方法影响，可能出现一些不匹配的情况。这个时候不防试试把网卡速度直接设为10M。其方法是右击“本地连接”打开其属性窗口，在“常规”选项卡中单击“配置”按钮，将打开的网卡属性窗口切换到“高级”选项卡，在“属性”列表中选中“Link Speed/Duplex Mode”，在右侧的“值”下拉菜单中选择“10 Full Mode”（图1），依次单击“确定”按钮保存设置。
无线网卡使用常见问题及解决方法 适用于型号：LB-LINK品牌的BL-LW05-AR5，BL-LW06-AR，BL-LW06-1R，BL-LW05-5R2，有购买此类型号的朋友敬请转载，保存，以便查用。 驱动安装 1、驱动的安装是先安装驱动还是先插网卡？安装显示微标认证怎么选择？ 安装驱动之前先把网卡插到电脑的USB口上。再把光盘打开，进行安装。 在安装驱动的时候电脑出现没经过微标认证，该如何选择？ XP系统出现的未经过微标认证，则点击继续安装。 WIN7系统出现的未经过微标认证，则点击始终安装此程序。 选择继续安装对电脑没有任何的损坏。
WIN7安全提示： 2、安装好驱动之后我在哪里找到驱动程序的配置？ 网卡安装好之后，我们需要对网卡进行相对应的配置。 如果你购买的是05-AR5，则在安装好驱动之后在电脑的桌面会出现一个绿色的图标，可以通过以下三种方式打开网卡的设置界面：
1）双击桌面的驱动图标，会弹出设置界面 2）双击屏幕的右下角的驱动图标 3）在开始----所有程序里面找到：BLINK 11N 的程序，选择即可打开。 3.我安装好驱动之后要不要重启电脑？ 一般来说如果电脑已经正确的识别了网卡，则不需要重启电脑也可以对网卡进行配置，如果你的电脑在安装好驱动之后还是不能识别网卡，则需要把电脑重启，然后再对网卡进行配置。 4、我安装好驱动之后在网卡的设置界面只有关于，刷新，和说明书的不一样。 这种情况就是属于网卡的驱动没有安装成功，如果你是在台式机上安装，请将网卡插到电脑的后面，然后把驱动卸载了，打开光盘，复制光盘内的windows driver文件夹到桌面，双击打开里面的setup.exe文件，即把驱动重新安装一遍。 做接收功能时候设置 5、怎么连接上已经有的无线信号？ 1）在打开网卡的配置程序后，找到”可用网络”,则网卡会自动的搜索附近的无线信号，如果你的无线信号在附近，就会被搜索到。