102 results about "Hyperspectral image processing" patented technology

The invention discloses a gastrointestinal tumor microscopic hyper-spectral image processing method based on a convolutional neural network, comprising the following steps: reducing and de-noising the spectral dimension of an acquired gastrointestinal tissue hyper-spectral training image; constructing a convolutional neural network structure; and inputting obtained hyper-spectral data principal components (namely, a plurality of 2D gray images, which are equivalent to a plurality of feature maps of an input layer) as input images into the constructed convolutional neural network structure using a batch processing method, and by taking a cross entropy function as a loss function and using an error back propagation algorithm, training the parameters in the convolutional neural network and the parameters of a logistic regression layer according to the average loss function in a training batch until the network converges. According to the invention, the dimension of a hyper-spectral image is reduced using a principal component analysis method, enough spectral information and spatial texture information are retained, the complexity of the algorithm is reduced greatly, and the efficiency of the algorithm is improved.

A method to jointly improve resolution of high spectral image space and spectral information relates to a method to improve spatial resolution through high spectral image information, which removes a failure to make full use of spatial information and spectral information to improve image resolution during high spectral image processing and comprises steps below: I. Inputting high spectral image data; A. Withdrawing spatial information; A I. Selecting characteristic wave band; A II. Analyzing and judging partial space; B. Withdrawing spectral information; B I. Withdrawing spectral terminal element; B II. Mixing pixel decomposition; II. Fulfilling collaborative super resolution of space and spectrum; III. Obtaining high spectral images with improved resolution. The present invention realizes breakthrough of spatial resolution during image acquisition, utilizes mixed partial relevance supporting vector mechanical decomposition to conduct spatial and spectral information collaboration technology, improve spatial resolution of high spectral image, greatly increase target detecting and locating capacity, break through limits of image acquisition means and make up hardware defects.

The invention belongs to the field of hyperspectral image fusion processing and specially relates to a hyperspectral image fusion method for hyperspectral image fusion and spatial resolution enhancement based on end member extraction and spectrum unmixing. The method comprises the steps of: using an N-FINDR algorithm to carry out end member extraction; using a spectrum unmixing technology to obtain an abundance value of each end member in each pixel; using an abundance matrix A as prior knowledge, carrying out classified marking on pixels of a plurality of spectral images by means of a fuzzy C mean value clustering algorithm, and then carrying out fused image reconstruction according to marking results and end member spectrums; obtaining classifying results, assigning end member spectrums to each pixel of a hyperspectral image according to marked categories, and obtaining a reconstructed fused hyperspectral image. According to the invention, the end member extraction technology is used for extracting and reserving end member spectrum information, no coefficient conversion steps are introduced in the whole fusion process, so that spectrum distortion is avoided; in addition, compared with an existing hyperspectral image fusion method, the hyperspectral image fusion method provided by the invention is better in spectrum fidelity.

The present invention discloses a Non-negative Matrix Factorization (NMF) based on sparse and correlation constraints and the method is applied to processing of decomposition of mixed pixels of a hyper-spectral remote sensing image. According to the method, finally, a given non-negative matrix Vm*n is factorized into a product of a basis matrix Wm*r and a coefficient matrix Hr*n, i.e. Vm*n is approximately equal to Wm*r Hr*n ; firstly, the non-negative matrix V is selected, W and H are randomly initialized, then the minimum correlation constraint is applied to the coefficient matrix H in a target function, the sparse constraint is applied to the basis matrix W and the coefficient matrix H and then iteration is carried out according to an iteration formula until the matrices W and H are converged.

The invention provides a method and a system for detecting food freshness in a refrigerator, and the method is as follows: acquisition of hyperspectral image information of food in a to-be-detected position in the refrigerator; uploading of the hyperspectral image information containing food freshness information to a server to obtain first shelf life information of the food by use of a hyperspectral image processing algorithm by the server. The first shelf life information of the food can be obtained by the hyperspectral image processing algorithm, so that the food freshness can be determined, the method has high accuracy, the method avoids to misjudge whether food the stored in the refrigerator is fresh only by experience, and unpredictable consequences caused by the misjudgment can be prevented.

The invention discloses an evolutionary multi-objective optimization-based method for extracting end members of hyperspectral remote sensing images, and belongs to the field of hyperspectral remote sensing image processing. According to the method, an end member number is taken as an objective function of multi-objective optimization, and different numbers of end member extraction results are obtained by adoption of a singly operation algorithm, so that the execution speed of the algorithm is improved and the precision is improved; extraction of end members of hyperspectral remote sensing images is considered as a multi-objective problem, two objective functions are optimized at the same time by utilizing a discrete particle swarm optimization method, and through single operation, different end member numbers can be obtained, namely, optimum end members can be obtained; and the problem that different numbers of end member results are obtained through executing the single operation algorithm for multiple times in the prior art is overcome. By adoption of a reverse-growth leader selection strategy, the users do not need to search all the end members, so that the calculation complexity is decreased.

The invention provides a clustered adaptive window based hyperspectral image abnormality detection method which belongs to the hyperspectral image processing field with the object of solving the problem with the consistence of the hyperspectral image background restricted by an existing background model structuring method. The steps of the method are as follows: conducting analysis on the main components of spectral dimensions of hyperspectral image and generating spectral subspace; generating adaptive windows for each to-be-detected pixel wherein each of the generated adaptive window is a binary matrix whose center is superposed with the to-be-detected pixel and the pixel in the matrix represented by one indicates the pixel as one in the homogeneous background area of the hyperspectral image while the pixel in the matrix represented by zero indicates the pixel as one in the non-homogeneous background area of the hyperspectral image; using the analysis result of the main components and an elliptical contour model to estimate the background logarithmic likelihood of the adaptive window to detect the abnormal image elements and generate a preliminary matrix for detection result; and using the morphological filtering for post-treatment and obtaining the final result of the detection matrix. The invention is used to detect the abnormity of a hyperspectral remote sensing image.

The invention provides a method and device for processing a hyperspectral image. The method comprises the steps that the hyperspectral image to be processed is determined; an abnormal target is extracted from the hyperspectral image to be processed; the abnormal target is separated to obtain an abnormal detection result. According to the method for processing the hyperspectral image, the hyperspectral image to be processed is determined, the abnormal target is extracted from the hyperspectral image to be processed, and the abnormal target is separated to obtain the abnormal detection result. Therefore, the abnormal target is extracted from the hyperspectral image and is separated to obtain the abnormal detection result, the false alarm rate can be lowered, and therefore locating precision is improved.

The invention provides a tensor decomposition cutoff remote sensing hyperspectral image compression method based on fast optimal core configuration search and relates to a hyperspectral image processing method. Aiming at the problem that a compression method based on the tensor decomposition cannot easily and fast obtain the optimal tensor core configuration under the requirement of setting the compression quality and the compression ratio, the tensor decomposition cutoff remote sensing hyperspectral image compression method based on the fast optimal core configuration search is provided. The method has the following steps that hyperspectral images are subjected to complete Tucker decomposition; spectrum dimension search starting points are calculated, iterative search is started, and the spectrum dimension optimal configuration is obtained; then, the trimming iteration is carried out, and the space dimension optimal configuration is obtained; and finally, complete decomposition results are intercepted, and final compression results are obtained. The method can be applied to satellite-bone or ground hyperspectral image compression, the compression recovery quality is ensured, and meanwhile, the calculation quantity of the compression method can be effectively reduced.

The invention provides a winter wheat yield prediction method based on unmanned aerial vehicle imaging hyperspectral remote sensing, and the method comprises the following steps: hyperspectral image processing of an unmanned aerial vehicle, growth parameter inversion of crops, construction of a winter wheat yield prediction model, and verification of yield prediction precision. According to the prediction method provided by the invention, wheat growth vigor information obtained through hyperspectral remote sensing of the unmanned aerial vehicle in multiple growth periods is comprehensively considered; meanwhile, crop growth priori knowledge provided by a crop growth model is introduced to predict the wheat yield, the method is simple in calculation and accurate in result, has universalitystarting from a remote sensing mechanism, and provides a new thought and a new method for accurately predicting the crop yield.

The invention proposes a wrapper-type hyperspectral waveband selection method based on pixel clustering. The method comprises the following specific operation steps: inputting a hyperspectral image for waveband selection, and converting the hyperspectral image into a matrix; carrying out the superpixel segmenting of hyperspectral data, and obtaining superpixel blocks; selecting a representative point from each superpixel block through employing a correlation method; firstly employing a non-supervision k-mediods method to achieve the clustering of all pixels, secondly employing an svm classifier for further optimizing a clustering effect, and obtaining a final clustering result; enabling the representative points to serve as a mark sample through employing the final clustering result, and employing a wrapper method to select wavebands. The method solves a technical problem that a supervision waveband selection method cannot be used when there is no mark sample. The method is wide in application range, is good in selection effect, employs the supervision waveband selection method in a non-supervision waveband selection field, and enlarges the application range of supervision waveband selection. The method is used for data dimension reduction in hyperspectral image processing, and facilitates the subsequent data processing.

The invention discloses a hyperspectral-analysis-based copper quality detection method, which comprises the following steps of: (1) constructing a characteristic image cube of a copper material to be detected; (2) extracting contents and spectral information of ingredients of the copper material to be detected from the characteristic image cube; and (3) evaluating the quality of the copper material to be detected. According to the method disclosed by the invention, the concept of the characteristic image cube is introduced based on the traditional hyperspectral image processing technology, meanwhile, the image characteristics and the spectral characteristics of the copper material to be detected are acquired, and the copper material is subjected to characteristic extraction and quality detection by using a fused PSO-nsNMF (Particle Swarm Optimization-based non-smooth Nonnegative Matrix Factorization) algorithm, therefore, the accuracy in the detection of copper quality is ensured. Theinvention further discloses a hyperspectral-analysis-based copper quality detection system which comprises a hyperspectral imaging unit and an image processing unit; the hyperspectral-analysis-based copper quality detection system disclosed by the invention is used for carrying out real-time online acquisition and data processing on an image of a copper sample by utilizing a computer and combining with a hyperspectral imaging instrument, therefore, the timeliness in the detection of copper quality is ensured.

The invention belongs to the technical field of hyperspectral image processing, and discloses a spectral super-resolution adaptive weighted attention mechanism deep network data processing method, which comprises the steps of selecting two spectral reconstruction challenge data sets for verification; evaluation index selection: using a root mean square error RMSE and an average relative absolute value error MRAE as evaluation indexes, and then calculating the MRAE and the RMSE; constructing an adaptive weighted channel attention mechanism module; constructing a second-order non-local module based on partitioning; and constructing an adaptive weighted attention mechanism network and training. According to the invention, a brand-new adaptive weighted attention mechanism network is designed for spectral super-resolution, and a backbone network is formed by stacking a plurality of double-residual modules which realize double-residual learning through long and short connection. And a self-adaptive weighted channel attention module is embedded in the double-residual module, so that the channel characteristic response is recalibrated, and the characteristic expression capability of the network is enhanced.

The invention relates to a hyperspectral image processing method for extracting the information of jarosite, wherein the wave band of jarosite, obvious in spectral features, is reserved. According to the method, the images of 980 nm, 1355 nm, 1415 nm, 1460 nm, 1550 nm, 1595 nm, 1790 nm, 2015 nm, 2075 nm, 2120 nm, 2225 nm, 2255 nm, 2330 nm and 2405 nm in wave band are extracted to be subjected to a series of judgment and calculation. Therefore, the abundance valves of jarosite in different areas can be calculated within the image range. Based on the above method, wave bands, which are not obvious in spectral features, are removed, so that the spectral features of jarosite can be emphasized during the information extraction process. The influence of other ground features or the noise can be relieved, and the to-be-processed data volume is reduced. Moreover, the information of a final result can be extracted based on the small amount of manual operation by utilizing the IDL program. As a result, the accuracy and the speed of the jarosite information extraction are improved.

The invention relates to a hyperspectral image processing method used for extracting calcite information. The method includes preserving the wavebands having significant spectral signatures of calcite, extracting images the wavebands of which are at 1760 nm, 1775nm, 1940 nm, 2000 nm, 2030 nm, 2105 nm, 2150 nm, 2165 nm, 2180 nm, 2195 nm, 2330 nm, 2345 nm and 2390 nm, performing a series of determination and calculation, and calculating abundance values of the calcite in different zones in the image scope. The method can remove the wavebands without significant signatures, thus highlighting the spectral signatures of the calcite in an information extraction process, reducing influences of other ground features or noises, and reducing the amount of data to be processed. Through an IDL program, the method can achieve an objective of information extraction of final results by less manual operation, thus increasing the precision and the speed of calcite information extraction.

A method of defining a software configuration for a hyperspectral imaging apparatus includes: creating a plurality of different functionalities, each based on at least one wavelength range; storing one or more of the created functionalities into a hyperspectral image processing model; generating a hyperspectral application for a specific hardware implementation of the hyperspectral imaging apparatus by configuring the hyperspectral application to utilize one or more of the created functionalities of the hyperspectral image processing model; and setting a selection of wavelengths as a control parameter in the hyperspectral application for controlling the specific hardware implementation of the hyperspectral imaging apparatus during an image capture. The selection of wavelengths is based on the wavelength ranges of the utilized one or more created functionalities.

The invention provides a hyperspectral image processing method based on feature strengthening. Textural feature matrixes composed of feature values of all pixels in two-dimensional images are used for describing the corresponding two-dimensional images; the textural feature matrixes of all the two-dimensional images are subjected to textural feature strengthening to obtain textural feature strengthened matrixes; main textural features are extracted according to all the textural feature strengthened matrixes to form a main textural feature vector; the main textural feature vector is used for representing a hyperspectral image. According to the hyperspectral image processing method based on feature strengthening, multi-wavelength information of the hyperspectral image is reasonably utilized, rich textural features can be captured accurately, texture details can be distinguished conveniently, and the hyperspectral image processing method is particularly suitable for the analysis of fine-texture images.

The invention relates to the technical field of hyperspectral image processing, in particular to a hyperspectral image classification method based on a flat hybrid convolutional neural network. According to the method provided by the invention, convolution of multiple dimensions is utilized; three-dimensional convolution is introduced into the first several layers of the primary neural network model; according to the method, spatial-spectral features are extracted and expressed, the latter several layers are connected with a two-dimensional convolution layer, and learned features are further integrated, so that the defects of large space occupation, time consumption and slow convergence of single three-dimensional convolution are avoided, and more effective features can be learned than single two-dimensional convolution; according to the method, pooling of various types is combined for sampling, learned effective features are utilized and reserved as much as possible while feature dimension acceleration training is reduced, model parameters are greatly reduced, the over-fitting phenomenon is relieved, the feature learning capacity can be kept under the condition of few training samples, and a good classification effect is obtained.

A novel hyperspectral image classification method based on wavelet packet transformation and a grey prediction model belongs to the hyperspectral image processing field. The method comprises the following steps: firstly, acquiring hyperspectral data to be processed; secondly, using the wavelet packet transformation to decompose a hyperspectral response curve of each pixel; thirdly, using the grey prediction model to process a decomposition result; fourthly, using a characteristic construction result to supervise and classify hyperspectral data; fifthly, outputting a hyperspectral image ground object classification result. The method is an automatic hyperspectral image classification method. By using the method, wave band correlation can be effectively removed; data redundancy can be reduced; a negative effect of a dimension disaster on classification precision can be avoided; an application range is wide.

The invention belongs to the technical field of hyper-spectral image processing, and especially relates to a hyper-spectral image nonlinear de-mixing method based on differential search. The method comprises the following steps: extracting end members of a real hyper-spectral image; determining the dimensionality and position code of each search individual; calculating the fitness value of each search individual according to an objective function; calculating a stop-over site i; comparing the fitness value of the current position Xi of each search individual with the fitness value of the stop-over site i thereof; and deciding whether to perform calculation according to conditions. According to the method, a nonlinear de-mixing problem is converted into an optimization problem, and the limitation that the traditional gradient optimization algorithm has high requirement on the initial value and easily falls into local convergence is overcome. Compared with a linear de-mixing algorithm and a gradient optimization-based nonlinear de-mixing algorithm, the hyper-spectral image nonlinear de-mixing method of the invention has the advantages of higher de-mixing precision and higher stability.

The invention discloses a hyperspectral image stripe missing restoring method based on an edge constraint and self-adaptive morphological filtering, belongs to the field of hyperspectral image processing in the remote sensing image processing, and aims to solve the problem of incapability of restoration when stripe missing occurs in continuous spectrum bands in the acquisition process of a hyperspectral image. The hyperspectral image stripe missing restoring method comprises the steps of detecting a stripe and determining a specific position of the stripe missing in the hyperspectral image; restoring the edge and preferably restoring edge information in the stripe missing; generating a self-adaptive structural element based on the edge constraint for each damaged pixel of stripe missing, the self-adaptive structural element being capable of protecting the hyperspectral image information and greater than the width of the stripe missing so as to ensure the structural element can cover an undamaged region; and performing self-adaptive morphological filtering and determining a final restoring value of the damaged pixels of the stripe mission. The invention is used for hyperspectral image restoration.

The invention discloses a hyperspectral image classification method based on sparse low-rank regression, and mainly solves a problem that the processing speed of a hyperspectral image in the prior art is low. The method comprises the steps: (1) reading the hyperspectral image data, and carrying out the mean filtering of the hyperspectral image data; (2) determining a training sample and a testing sample in a spectrum component with a label in the hyperspectral image after filtering; (3) solving a low-rank projection matrix and a parameter matrix according to the training sample; (4) solving an embedded characteristic matrix of the training sample and an embedded characteristic matrix of the testing sample according to the low-rank projection matrix and the parameter matrix; (5) employing a linear supporting vector machine classifier to classify the embedded characteristic matrix of the training sample and the embedded characteristic matrix of the testing sample, and obtaining a classification image. The method is high in classification precision, is low in cost of high-dimensional data processing, and can be used for the discrimination of surface features of the hyperspectral image.

The invention relates to the field of hyperspectral image processing, in particular to a hyperspectral image classification method and device based on space spectral dimension filtering. According tothe method and the device, TSG filtering and black and white mask calibration are carried out on a hyperspectral image of a sample after reflectivity inversion; and constructing a feature set based onthe label information of the hyperspectral image of the sample and the first multiple principal components of the hyperspectral image of the sample, inputting the feature set into a training set to train a support vector machine, and classifying a test set by using the trained support vector machine. According to the method and the device, the hyperspectral image is constructed by combining principal component analysis and a support vector machine algorithm, so that DN value distortion caused by the influence of the three-dimensional form of a sample in the hyperspectral image can be inhibited, meanwhile, the strip noise of the spectral dimension of the image is improved, and the spatial-spectral dimension filtering of the hyperspectral image is realized. According to the method, DN valuedistortion caused by sample edges and irregular surfaces in the hyperspectral images is improved, the classification precision of the images is effectively improved, and the method and device can beapplied to the fields of agriculture, pharmaceutical industry, environmental monitoring and the like.

The invention discloses a hyperspectral image classification method, device and equipment and a storage medium, and belongs to the technical field of hyperspectral image processing. According to the invention, the method comprises the steps: acquiring a hyperspectral image; performing space conversion on the hyperspectral image through a convolutional neural network; obtaining convolutional images, and segmenting the convolution image; obtaining a plurality of segmented image segments; performing convolution operation on each image segment to obtain a convolution segment, connecting the convolution segments to obtain a target convolution segment, inputting the target convolution segment to a full connection layer in the convolutional neural network, and acquiring a picture classification result output by the full connection layer so that accurate classification of the high-dimensional hyperspectral image can be realized.