44 results about "Non-negative matrix factorization" patented technology

The invention discloses an intelligent multi-gas detection module data processing system and method based on NDIR, and relates to the technical field of gas detection.The method comprises the steps that a multi-wavelength NDIR sensor is used for obtaining light intensity changes of a gas sample under different wavelengths, and original spectral signals are generated; performing denoising processing on the spectral signal by adopting wavelet transform, and performing zero calibration and dynamic baseline deduction; based on the Beer-Lambert law in combination with environmental factors, establishing a relation model between gas absorption and spectral signals; updating model parameters by using a recursive least square algorithm, and continuously optimizing gas concentration prediction; separating gas signals by adopting a non-negative matrix factorization algorithm, and predicting the concentration of each gas; and the edge end operates the lightweight model in real time, and regularly uploads the model to the cloud end for global optimization and federated learning. The method can effectively improve the condition that the precision is insufficient when the model is used for a long time in the prior art.

The invention relates to the technical field of building energy saving, and discloses a building energy-saving control system and method.The method comprises the steps that firstly, multi-source environment data inside and outside a building is collected, and a time-space fusion data set is generated through timestamp synchronization and data alignment; dynamic energy consumption characteristics are extracted, and the future energy consumption state of each region is predicted by using a hidden Markov model; decomposing a historical energy consumption mode according to a prediction result and a non-negative matrix factorization algorithm to generate a multi-objective optimization control strategy; and finally, adjusting an equipment operation sequence through a dynamic priority scheduling algorithm and outputting an instruction. The system comprises a multi-source data acquisition module, a feature extraction module, a state prediction module, a strategy generation module, an equipment scheduling module and a feedback correction module. According to the method, energy consumption can be accurately predicted, multi-target optimization control, dynamic equipment scheduling and online strategy correction are realized, the building energy utilization efficiency is effectively improved, energy consumption is reduced, the comfort level is guaranteed, and equipment loss is reduced.

The invention discloses a cross-multi-camera multi-target tracking method, system, device, medium and product, and relates to the field of computer vision, and the method comprises the steps: obtaining a video image, employing a single-camera multi-target tracking architecture according to the video image, and obtaining a local track of single-camera multi-target tracking, the single-camera multi-target tracking architecture comprises a detector, a ReID module and a tracker; according to all single-camera multi-target tracking local tracks, a local track set is obtained, the target appearance features, the target detection frames and the target positions of every two local tracks and the topological information of the corresponding cameras are utilized to determine the track similarity measurement of the two local tracks, and then a matching method of limited non-negative matrix factorization is adopted to determine the target similarity of the two local tracks. Obtaining a global trajectory of the target; according to the invention, a high-precision tracking task of a target in a complex scene can be realized.

The invention provides a face image clustering method and device, equipment and a storage medium. Relates to the technical field of image processing. The method comprises the following steps: inputting required clustering data, wherein the clustering data is a high-order graph regular non-negative matrix; calculating and constructing a high-order similarity matrix; constructing a high-order graph constraint; defining and initializing each layer of basis matrix and representation matrix of regular non-negative matrix factorization of the high-order map; obtaining an updating formula of each layer of basis matrix and an expression matrix of the regular non-negative matrix factorization of the high-order map; updating the basis matrix and the representation matrix of each layer of the regular non-negative matrix factorization of the high-order map; obtaining a low-dimensional representation matrix; and outputting a clustering result. According to the invention, for the face image data with complex similarity, similarity information of more levels can be captured, so that the clustering performance is improved.

A method for operating a driver monitoring system for a vehicle, comprising the following steps: Receiving a plurality of gaze target points for a driver of the vehicle; inputting the plurality of gaze target points into a predictive distraction probability model to obtain a predictive distraction distribution; Determine whether one or more meaningful gaze locations exist in the plurality of gaze target points; Comparing the predictive distraction distribution to a predictive distraction distribution threshold if one or more meaningful gaze locations are present in the plurality of gaze target points; and Warn the driver when the predictive distraction distribution reaches or exceeds the predictive distraction distribution threshold; wherein the predictive distraction probability model includes developing a graphical representation of the plurality of gaze target points; wherein the graphical representation is a histogram graph of the plurality of gaze target points; wherein the graphical representation is used to develop a gaze gray-level matrix that captures a plurality of gaze transitions with the plurality of gaze target points; wherein the method further comprises the step of factoring the gaze gray value matrix using a non-negative matrix factorization (NMF) technique to develop a plurality of NMF factors.

The invention provides a wheat leaf area index whole growth period estimation method based on canopy pure hyperspectral reflectivity, which comprises the following steps of: firstly, extracting wheat canopy spike, leaf and soil pure pixels based on an RGB (Red, Green and Blue) image combined with a deep learning algorithm; secondly, separating the canopy pure leaf hyperspectral reflectivity by using the extracted pure pixels and canopy hyperspectral reflectivity data in combination with a non-negative matrix factorization algorithm; and finally, calculating a vegetation index based on the canopy pure leaf hyperspectral reflectivity, and establishing a whole-growth-period LAI estimation model in combination with a simple linear regression algorithm. According to the method, the problem that the LAI inversion precision is reduced after heading is solved by combining near-end remote sensing and a machine learning algorithm, and the wheat LAI estimation method based on the canopy pure hyperspectral reflectivity has the potential of accurately predicting the LAI of the wheat in the whole growth period and can be expanded to accurate management of field crops and breeding of high-yield wheat varieties.

This application discloses a deconvolution method for gas chromatography-mass spectrometry (GC-MS) data, relating to the field of GC-MS data analysis technology. This method utilizes a non-negative matrix factorization function to initially decompose the time-mass-charge ratio two-dimensional intensity matrix, obtaining an initial elution curve matrix and an initial mass spectrometry matrix. An exponentially modified Gaussian peak shape modeling function is then used to fit the initial peak shape parameter set. For each component: based on the initial peak shape parameter set, an exponentially modified Gaussian peak shape modeling function is used to generate the original elution curve matrix. Non-negative least squares optimization and non-linear least squares optimization methods are used to iteratively optimize the mass spectrometry matrix, peak shape parameter set, and elution curve matrix. Finally, a comprehensive scoring function is used to select the optimal component set, and the optimal peak shape parameter set, mass spectrometry matrix, and elution curve matrix for the optimal component set are sorted and output. This application enables precise separation of co-eluted components, improving the accuracy and reliability of complex mixture analysis.

The application proposes an unsupervised explanation method based on adaptive semantic features, and the constructed explainable model mainly includes three core modules: feature-level explainability module, adaptive feature expression module and feature importance calculation module. Firstly, the deep feature mapping of the deep neural network is regarded as the abstract expression of the high-level concept semantics learned by it, then the non-negative matrix decomposition technology is used to unsupervisedly extract key semantic information from it, and feature-level explanation or structured attribution is carried out; through the dimension scaling scheme, adaptive number of feature extraction is realized; in the feature importance calculation, the Shapley value algorithm is used for calculation. In addition, a saliency visual explanation is generated to highlight the key area of the model decision. Experiments show that the method has higher explanation accuracy, and in the environment of different data sets and explained models, the explanation accuracy is better than that of the existing method, and has the characteristics of accurate explanation, robustness and universality.

The invention relates to the technical field of immunohistochemical staining image quantitative analysis, in particular to an immunohistochemical staining image quantitative analysis method and system, and the method comprises the following steps: S1, obtaining a digital image of an immunohistochemical staining section; and S2, carrying out color deconvolution processing on the digital image, and decomposing the digital image into an exclusive channel image corresponding to at least one target coloring agent and a background channel image. According to the method, color deconvolution is realized through a non-negative matrix factorization algorithm, target coloring agent signals and background and other dyeing signals in the composite dyeing image are accurately split, and the problem that dyeing components of an existing system are not thoroughly separated is solved; through combination of a Mask R-CNN deep learning network and multi-mode adaptive parameters, positive and negative cells in different dyeing modes can be accurately identified, and multiple scenes such as nuclear dyeing, membrane dyeing and slurry dyeing can be adapted.

The invention relates to the field of machine learning and pattern recognition, and particularly discloses a variance-covariance subspace distance-based multi-kernel subspace clustering method, which comprises the following steps of: S1, acquiring a high-dimensional data matrix, calculating a variance-covariance matrix and second-order statistical information thereof, defining a preliminary VCSD kernel value and obtaining a VCSD kernel matrix through exponential transformation and normalization, constructing a multi-core pool of r base cores; s2, constructing an optimized objective function based on non-negative matrix factorization, and solving to obtain a local affinity feature map of a block diagonal structure under constraint conditions; s3, capturing high-order feature association through a matrix method, and converting tensor optimization into matrix operation; s4, solving the target function by adopting an alternating optimization strategy, reducing the calculation complexity, and generating a target affinity graph; and S5, clustering the target affinity spectrum, and outputting a result. According to the technical scheme provided by the invention, the problems of missing statistical characteristics, complex calculation and nuclear noise interference of the existing method are solved, and the clustering precision and efficiency are effectively improved.

The application belongs to the field of satellite data processing, and particularly relates to a Swarm satellite electron density data daily global drift disturbance automatic identification and removal method, which comprises normalization processing; short-time Fourier transform is performed on the normalized data to obtain a time-frequency matrix, the time-frequency matrix is classified into strong disturbance orbit data and other orbit data by using a two-dimensional convolutional neural network, and a component containing daily global drift disturbance is identified; for the decomposition result of the strong disturbance orbit, the identified disturbance component is removed; for the disturbance component of the other orbit, a medium-strong disturbance interval is identified and located, and a disturbance latitude range is obtained according to a mapping relationship between an image coordinate and a latitude; for the identified medium-strong disturbance orbit, pre-normalization data thereof are obtained and subjected to non-negative matrix decomposition, a corresponding part in the disturbance component is cut off according to the disturbance latitude range, and the remaining part and the remaining components are added to serve as effective signals and reserved, so that automatic cleaning of massive satellite data is realized.

The invention relates to the field of machine learning, provides an artificial intelligence-based occupational hazard risk assessment system, and aims to solve the problems of serious missing of multi-dimensional risk data, insufficient feature expression and low assessment precision. The system is composed of a data acquisition module, a data enhancement module and a risk assessment module. Unified tensorization processing is carried out on the data through a segmented weighted non-negative matrix factorization model, an error function of L1 and L2 norm combination is introduced, and the robustness and accuracy of data completion are improved; on this basis, constructing a Bargmann-SVM model, mapping standardized data to a Bargmann space, obtaining complex high-dimensional features by extracting a high-order derivative and a projection coefficient of a full-pure function, training a support vector machine in combination with a polynomial radial basis kernel, and performing accurate classification and grade evaluation; the system can be widely applied to high-risk environments such as factories and workshops, and the intelligent recognition and early warning capability of occupational hazard risks is improved.

The invention discloses a non-negative matrix factorization face clustering method and device based on graph learning and pseudo tag guidance and a medium, and relates to the technical field of data analysis and machine learning. The method comprises the following steps: acquiring a data matrix, a pseudo tag matrix, a regularization parameter and a clustering number; the data matrix comprises face image data; randomly initializing a basis matrix, a coding matrix, a self-expression coefficient matrix and an auxiliary matrix; determining an affinity matrix and a diagonal matrix, and determining a graph Laplacian matrix according to the affinity matrix and the diagonal matrix; and carrying out iteration based on the set maximum number of iterations, and in each iteration process, sequentially updating each matrix and minimizing the target function until the target function converges or the number of iterations reaches the maximum number of iterations, so as to obtain a clustering result. According to the method, by combining a self-representation method in subspace learning, the representation coefficient is generated, the Laplacian matrix is further calculated, and the Laplacian matrix is updated in the iteration process, so that generation of the optimal solution is guaranteed.

The invention relates to a model-driven deep learning multicolor fluorescence blind source unmixing method, which comprises the following steps of: establishing an optimization problem of regularization constraint containing image denoising based on a linear spectrum mixing model, and constructing a non-negative matrix factorization blind source unmixing framework; non-negative matrix factorization is realized based on gradient descent and an iterative shrinkage threshold algorithm, the gradient descent and the iterative shrinkage threshold algorithm are expanded into a multi-layer recurrent neural network, and a transformation function is replaced by the multi-layer convolutional neural network; designing a spectrum extraction algorithm to obtain prior information of a fluorescence emission spectrum, and initializing parameters of the multilayer recurrent neural network according to the prior information; and constructing a multicolor fluorescence imaging simulation data set, and carrying out end-to-end training and testing on the multilayer recurrent neural network. Compared with a method for demixing the multicolor fluorescence image directly through a traditional non-negative matrix factorization algorithm, the method has the advantages that a better demixing effect is obtained under the condition that a small number of network layers is used, artifacts are fewer, and structural information is more accurate.

The invention relates to a community detection method and device and a storage medium, and the method comprises the steps: determining the harmonic centrality of each user in a communication user network, constructing community division according to the harmonic centrality and a preset community number estimation range, and obtaining a plurality of candidate communities; calculating the affiliation probability and the community modularity of nodes in each candidate community so as to determine the optimal community number; selecting a target community center node according to the optimal community number and the harmonic centrality of each user; constructing a node similarity matrix according to the target community center node, and constructing an optimization target function according to the node similarity matrix; and performing iterative solution on the target function by adopting a non-negative matrix factorization iterative optimization algorithm, and determining a community division result according to a solution result. The problem that the number of communities needs to be preset in a traditional method is effectively solved, and the method has good adaptivity and stability.

The application discloses a kind of electroencephalogram emotion signal recognition method based on graph regular non-negative matrix decomposition, comprising the following steps: step one, using nearest neighbor method for electroencephalogram emotion signal constructs an adjacent matrix;Step two, establish a graph regular non-negative matrix decomposition model, and the non-negative constraint of matrix after decomposition is carried out;Step three, introduce a projection matrix, further to matrix implement three decomposition;Step four, the model is optimized;Step five, using the new representation PX obtained after the multiplication of trained model parameter P and sample X replaces the representation matrix V of original model, joins corresponding label matrix and trains a classifier in linear SVM classifier;Step six, the class of test sample is predicted by inputting into classifier.The application can effectively identify electroencephalogram emotion signal, compared with other classic electroencephalogram emotion recognition method, the application effectively improves recognition rate.

To provide a sound quality improvement device, a sound quality improvement method and a program, which can improve sound quality of sound separated and extracted from an acoustic signal in which a plurality of sounds are mixed.SOLUTION: A sound quality improvement device 1 includes a separation unit 10 and an estimation unit 20. The separation unit 10 performs independent component analysis on first acoustic signals SA1(t) and SB1(t), which are obtained by recording composite sound of sounds emitted from a plurality of sound sources (a first sound source O1 and a second sound source O2) at different positions (microphones 2A and 2B) and separates the signals into second acoustic signals SA2(t) and SB2(t) for each sound source. The estimation unit 20 performs non-negative value matrix factorization on the first acoustic signal SB1(t) and the second acoustic signal SA2(t) and estimates a third acoustic signal S3(t) corresponding to the sound source on at least one sound source in the plurality of sound sources (first sound source O1 and second sound source O2).SELECTED DRAWING: Figure 1

The present application provides a facial image clustering method, apparatus, device and storage medium. It relates to the field of image processing technology. The method includes: inputting the required clustering data, which is a high-order graph regular non-negative matrix; calculating and constructing a high-order similarity matrix; constructing a high-order graph constraint; defining and initializing the basis matrices and representation matrices of each layer of the high-order graph regular non-negative matrix decomposition; obtaining the update formulas of the basis matrices and representation matrices of each layer of the high-order graph regular non-negative matrix decomposition; updating the basis matrices and representation matrices of each layer of the high-order graph regular non-negative matrix decomposition; obtaining a low-dimensional representation matrix; and outputting the clustering results. The present application is aimed at facial image data with complex similarities, and can capture more levels of similarity information, thereby improving clustering performance.

The application discloses a neural anatomical heterogeneity analysis method integrating standard model and non-negative matrix factorization, and belongs to the field of biomedical signal processing. The application comprises: preprocessing structural magnetic resonance imaging data of patients and corresponding healthy control groups to extract brain morphological indexes; establishing a brain morphological standard model of the patient through a transfer pre-training model to estimate a deviation score matrix of the patient's cortical thickness and surface area, and extracting a non-positive part to obtain a brain morphological atrophy deviation matrix; performing non-negative matrix factorization on the atrophy deviation matrix to generate a preliminary disease factor matrix and a weight composition matrix; and rearranging and sampling the deviation matrix, calculating an average half-decomposition stability coefficient and a reconstruction error to determine the disease factor and the weight composition under the optimal component number. The method of the application integrates spatial pattern information at the group level while retaining individual heterogeneity information, and provides a new technical means for precise diagnosis of neuropsychiatric diseases.

The invention belongs to the technical field of image processing, and particularly relates to a dietary tracking image processing method based on non-negative matrix blind source separation, which comprises the following steps: acquiring a dietary image, and preprocessing the dietary image; inputting the preprocessed image into a Unet network for food image classification; analyzing the classified food images by adopting a non-negative matrix factorization algorithm to obtain different types of food images; converting different types of food images into an HSV color space, and performing thresholding processing on the food images converted into the HSV color space; obtaining an independent food image according to the image after thresholding processing; calculating volumes of the individual food images; determining the nutrient intake of the user according to the volume of the food image and the type and number of the food image, and realizing diet tracking of the user; according to the method, spatial constraint non-negative matrix blind source separation is introduced, the local structure and spatial correlation of the image are reserved, stacked or semi-covered food materials with similar colors / textures are effectively distinguished, and the accuracy of component separation in a mixed scene is remarkably improved.

The invention discloses a robust non-negative matrix factorization hyperspectral image demixing method. The method comprises the following steps: step 1, calculating a noise estimation value and a noise estimation matrix of a hyperspectral image; step 2, constructing an improved weighted residual non-negative matrix factorization model based on the noise estimation value and the noise estimation matrix; 3, introducing a second-order adjacency relation to obtain a high-order graph, and carrying out Laplacian regularization to obtain a regularization term; 4, introducing a high-order graph Laplacian regular term to obtain a robust non-negative matrix factorization model based on global structure constraint; and step 5, realizing hyperspectral image unmixing based on the robust non-negative matrix factorization model of the global structure constraint. According to the method, a more robust and more accurate unmixing effect can be obtained, the internal structure of the hyperspectral image can be better protected, and the method is more suitable for scenes with more complex ground features, richer details or higher unmixing precision requirements.

This invention belongs to the field of computer data clustering technology and discloses a multi-view clustering method, apparatus, electronic device, and storage medium. The method includes: acquiring multi-view data to be clustered, wherein the multi-view data is one of images, videos, audio, and text; determining the Laplacian matrix, depth decomposition matrix, and global consensus embedding matrix for each view based on the original data matrix of the multi-view data; performing hypergraph regularized deep non-negative matrix decomposition on each layer of the original data matrix through the depth decomposition matrix to obtain the decomposition result; performing high-level semantic alignment and consensus representation fusion on the decomposition result through the global consensus embedding matrix to obtain the final layer representation of the view; determining the clustering indicator matrix of the final layer representation according to the objective function; and processing the clustering indicator matrix using the k-means clustering algorithm to obtain the multi-view clustering result. Through the technical solution of this invention, the accuracy of multi-view clustering is achieved.

The present invention belongs to the technical field of radar anti-jamming, and particularly relates to an anti-radar main lobe repeater jamming method based on a reversible residual network. Based on the non-negative matrix factorization (NMF) algorithm and the reversible residual neural network, the present invention proposes a deep learning framework Inv-NMF net suitable for suppressing radar main lobe repeater jamming in a multi-target and multi-jamming environment. It can use the radar transmitted signal information to decompose the jamming signal, and represent the target signal information and the jamming signal information at different positions in the coefficient matrix. Then, the reversible residual network is used to analyze the coefficient matrix to generate a Mask mask matrix, and separate the jamming signal component and the target signal component. Due to the lossless information transmission of the reversible network, the present invention can ensure that the target signal strength is not affected to the greatest extent. This method is not only not affected by the number of radar targets, but also has good adaptability to different radar transmitted waveforms.

The application particularly relates to a simulation method of a completely non-stationary wind field of a mountain bridge, which comprises the following steps: obtaining an evolution power spectrum function of a completely non-stationary wind field of a bridge to be simulated; determining representative interpolation nodes based on the evolution power spectrum function; the interpolation nodes comprise time domain interpolation nodes and frequency domain interpolation nodes; performing Cholesky decomposition on the interpolation nodes to obtain corresponding node Cholesky decomposition values; then further decomposing the node Cholesky decomposition values into a series of time and frequency function products by a non-negative matrix decomposition method; applying a Hermite interpolation method to establish global time and frequency interpolation functions; and generating a simulation high-efficiency calculation expression of the corresponding completely non-stationary wind field of the bridge based on the global time and frequency interpolation functions, so as to realize the simulation of the completely non-stationary wind field of the mountain bridge. The simulation method in the application can adapt to the completely non-stationary wind field, can take into account the time-varying frequency characteristics and spatial correlation of the wind speed field simulation, and can improve the practicability and efficiency of the bridge wind field simulation.

The present invention relates to an electromagnetic wave rain attenuation completion measurement method based on non-negative matrix decomposition, which belongs to the field of wireless communication. The method first uses a linear regression model to preliminarily fill in the missing values ​​in the matrix, then uses the data of other known columns as features to fit the linear regression model, and uses the model to predict the missing values ​​to improve the accuracy of initialization, and finally performs non-negative matrix decomposition on the initialized matrix, and by traversing different potential dimensions, selects the dimension value that can maximize the correlation coefficient of non-missing values ​​between the reconstructed matrix and the measurement matrix as the final result. The method of the present invention does not rely on a specific rain attenuation model or parameter setting, has strong versatility, and can adapt to the needs of completing electromagnetic wave rain attenuation data in different regions and under different parameter conditions.

The present application provides an adaptive graph clustering method and device based on robust symmetric non-negative matrix decomposition, wherein the method comprises: obtaining a target data set, wherein the target data set includes a plurality of sample vectors; calling a preset objective function of adaptive graph clustering based on robust symmetric non-negative matrix decomposition, clustering and dividing the plurality of sample vectors in the target data set, and obtaining a plurality of groups of clustering results of the plurality of sample vectors; processing the plurality of groups of clustering results through integrated clustering, and obtaining an optimal clustering result of the target data set. By utilizing the preset objective function of adaptive graph clustering based on robust symmetric non-negative matrix decomposition to better reflect the local geometric structure of the data, the robustness of the algorithm is enhanced, and the integrated clustering idea is combined with robust symmetric non-negative matrix decomposition to improve the algorithm clustering performance and adaptability to data of different structures, thereby improving the accuracy of the clustering results.

The application provides a mechanical and electrical equipment shock absorption and noise reduction installation method, and belongs to the technical field of mechanical and electrical equipment installation. A multi-dimensional vibration noise collection system is first constructed to collect comprehensive vibration noise characteristic data. Then, vibration noise characteristics are extracted through wavelet transform and non-negative matrix decomposition. Next, a small-scale model is constructed, and the influence of installation parameters is analyzed by using an orthogonal test method. Singular value decomposition is used to obtain a vibration noise variation matrix. A multi-objective optimization model is constructed based on this, and a particle swarm algorithm is used to solve optimal installation parameters. The rationality of the parameters is verified by using an elastic damping dynamics equation set. A matched shock absorber material and installation position are selected. Actual installation is performed according to the optimal parameters, and the effect is verified. Finally, a parameter correction model is established, a complete knowledge base is formed, and precise optimization of mechanical and electrical equipment shock absorption and noise reduction installation is realized.

The invention discloses a task state recognition system and method in an industrial agent for multiple industrial chains, and the system at least comprises a task feature extraction module and an incomplete multi-view clustering module, the task feature extraction module constructs a shared potential subspace through a non-negative matrix factorization technology based on task attributes and topological features, and the incomplete multi-view clustering module is used for clustering the potential subspace; complementing the missing task data, and generating a basis matrix of each view and potential representations of all views; and the incomplete multi-view clustering module is used for solving the optimal potential representation and supplementing complete task attribute data for the multi-view task data with missing data and the topological connection relationship through an incomplete multi-view clustering algorithm based on the base moment and the potential representation generated by the task feature extraction module, so as to realize the recognition of the task state. According to the system and the method disclosed by the invention, the task state can still be accurately judged by virtue of the clustering result under the adverse condition that part of attribute data of the task is missing, so that stable and efficient operation of multiple industrial chains is powerfully guaranteed.

The present application relates to a kind of based on hypergraph and multi-task cooperation small sample multi-pose face recognition method, belong to artificial intelligence face recognition technical field.The present application utilizes hypergraph and non-negative matrix decomposition to obtain image similar to frontal image, designs a kind of multi-pose face recognition framework based on hypergraph deflection.The framework is first separated to no pose deflection image.On this basis, a kind of feature coding method based on improved support vector description is proposed, the feature of no pose deflection image is extracted, and is optimized with the classifier based on dictionary learning, for feature extraction and feature classification.Feature coding method utilizes improved support vector data description and triangular coding, so that the feature extracted is more discriminative.At the same time, an effective feature extraction and feature classification optimization model is established, easy to obtain more close to global optimal solution, improves the recognition performance.

The present invention discloses a plug-and-play hyperspectral image unmixing method based on non-negative matrix decomposition. The present invention jointly considers manually selected regularization terms and regularization terms obtained through learning, and then utilizes prior information in hyperspectral images. The method uses a denoiser to learn spectral and spatial information, avoiding manually designing regularizers and solving complex optimization problems. The present invention also uses manual regularization terms to introduce some physically clear priors in the image. Taking sparse regularization terms as an example, the present invention adds prior knowledge of abundance sparsity, illustrating that the combined use of designed regularization terms and learned regularization terms can effectively improve the unmixing accuracy. The denoiser in the present invention can not only introduce prior information, but also the denoising characteristics of the denoiser improve the robustness of the algorithm to noise.