57 results about "Hyperspectral image compression" patented technology

The invention discloses a hyperspectral image compression method based on deep learning. The method comprises the following steps: firstly, selecting a training image, randomly cutting the size of thetraining image into a 32 * 32 size as a training set, and inputting the training set into a built compression network model for training to obtain a compression model comprising an encoding network,a quantization network and a decoding network; inputting an image to be compressed into a coding network, a coded feature map is obtained according to a calculation result of the coding network, theninputting the obtained feature map into a quantization network for quantization calculation to acquire code streams; finally inputting the quantized result into a decoding network, and acquiring a reconstructed image through calculation of a decoding network model. According to the invention,a hyperspectral image can be compressed, and the compression effect is improved.

The invention relates to a method for compressing an image, in particular to a method for compressing a high spectrum image. The method solves the problem that the lossless compression in the prior art cannot make a specific region or the image information of the specific region free from loss. The method is used for compressing an image. The method is characterized by comprising the following steps of: carrying out grade compression processing on an image to be compressed; carrying out lossless or nearly lossless compression on interested space information and interested spectral information; and compressing other information in which the compression rate is greater than that of the interested space information to the interested spectral information.

The invention discloses a hyperspectral image compressed sensing method based on the heavy weighting laplacian sparse prior. The hyperspectral image compressed sensing method based on the heavy weighting laplacian sparse prior is used for solving the technical problem that an existing hyperspectral image compressed sensing method is low in reconstruction accuracy. According to the technical scheme, a little linear observation of each pixel spectrum is collected randomly as compressed data, a compressed sensing model based on the heavy weighting laplacian sparse prior and a sparse regulated regression model is established, and solving on the established models is conducted. According to the hyperspectral image compressed sensing method based on the heavy weighting laplacian sparse prior, and a little linear observation of each pixel spectrum is collected randomly as compressed data, so that resource consumption in the image collecting process is reduced; the strong sparsity of the hyperspectral image is depicted accurately through the heavy weighting laplacian sparse prior, inhomogeneous constraint on the nonzero element of the traditional laplacian sparse prior is overcome, and the reconstruction accuracy of the hyperspectral image is improved. It is tested that when a sampling rate is 0.15 and the compressed data consist strong noise with 10 db signal-to-noise ratio, and the peak signal-to-noise ratio promotes over 4 db relative to a background technology method.

The invention discloses a hyperspectral unmixing compressive sensing method based on three-dimensional total variation sparse prior. The hyperspectral unmixing compressive sensing method is used for solving the technical problem that an existing hyperspectral image compressive sensing algorithm in combination with spectrum unmixing is low in precision. According to the technical scheme, a random observation matrix is adopted for extracting a small number of samples from original data as compression data. In the reconstruction process, according to an unmixing compressive sensing model, appropriate spectrums are selected from a spectrum library as an end member matrix in the model, then the three-dimensional total variation sparse prior of an abundance value matrix is introduced, and the abundance value matrix is accurately solved through solving a limited linear optimization problem. Finally, a linear mixing model is used for reconstructing the original data. When the compression ratio of urban data shot through a HYICE satellite is 1:20, the normalize mean squared error (NMSE) is smaller than 0.09, when the compression ratio is 1:10,the NMSE is smaller than 0.08, and compared with an existing compressive sensing algorithm, precision is promoted by more than 10%.

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 relates to hyperspectral image compressing and restructuring device and method. The method includes steps of firstly, subjecting each spectrum range of a hyperspectral image to three-level two-dimensional 9/7 integer discrete and lifting wavelet transformation; organizing wavelet coefficients to sections and blocks through CCDSD-IDC, encoding section header information and DC (direct current) coefficient initializing and AC (alternating current) coefficient depth of each block, extracting AC coefficient and residual DC coefficient bit plane simultaneously, and finally subjecting the AC coefficient and residual DC coefficient bit plane to distributed signal source encoding by a Slepian-Wolf encoder based on LDPC (low density parity check). The hyperspectral image compressing and restructuring device and method have the advantages that by adopting distributed signal source encoding strategy, complexity of the integral encoder is lowered, and the method is easy to implement through processors such as FPGA (field programmable gate array) and DSP (digital signal processor).

The invention provides a hyper-spectral image compression and encoding method and device. The hyper-spectral image compression and encoding method comprises the steps of detecting hyper-spectral images and obtaining target portions and background portions included in the hyper-spectral images; separating the target portions and the background portions; compressing and encoding the target portions and the background portions respectively. By compressing and encoding the target portions and the background portions respectively, the reliability and the high efficiency of hyper-spectral image compression and encoding are improved.

The invention discloses a hyperspectral image lossless compression method based on deep learning to solve the problems of inadequate utilization of spectral information and the low model generalization ability in conventional methods. The method comprises the following steps: establishing a prediction model by using a circulating neural network in depth learning; predicting and training each pixelin a hyperspectral image to generate a predicted image and a predicted network; subtracting the hyperspectral image from the predicted image and forming a residual image; performing arithmetic codingon the residual image and generating a code stream file; decoding the code stream file and obtaining a decoded image; performing prediction through a trained network and obtaining a predicated image;adding the predicted image and the decoded image and obtaining an original hyperspectral image. Deep learning and conventional methods are combined, units with a memory structure in the network are adopted, a lot of training is performed, spectrum information is fully utilized, the generalization ability of the model is improved, and the compression efficiency is improved. The hyperspectral imagelossless compression method is applied to the field of hyperspectral image compression.

The invention discloses a hyperspectral image lossless compression method based on the RKLT and principal component selection and belongs to the technical field of remote sensing hyperspectral image compression. The hyperspectral image lossless compression method based on the RKLT and principal component selection solves the problem that when an existing KLT method is used for hyperspectral image lossless compression, generated floating point number coefficients are not favorable for processing on hardware. According to the technical scheme, a hyperspectral image is converted into a 2D matrix from a 3D matrix; a transformation matrix is decomposed into four integer matrixes and transformation coefficients through the RKLT; RKLT reverse transformation is conducted on principal components which are selected from the transformation coefficients; subtracting is conducted on the matrix obtained after reverse transformation and the original 2D matrix, so that a residual error is obtained; predicting, forward mapping and section coding are conducted on the residual error and an RKLT forward transformation matrix of the selected principal components, so that a coding stream is formed; the transformation matrix generated by the KLT is stored into an RAW file, and the RAW file and the coding stream obtained in the last step serve as compressed data to be transmitted to a compression end; the number of the optimal principal components needing to be selected is found through a searching method. The hyperspectral image lossless compression method is suitable for conducting lossless compression on the hyperspectral image.

The invention discloses a hyperspectral image compression method based on non local and low-rank decomposition. Non local similarity is used for partitioning the hyperspectral image, hyperspectral small blocks with the same size are acquired; and then, according to the similarity, similar blocks are gathered in one class, four-order data formed by blocks in the same class can be subjected to low-rank decomposition. As the size of the decomposed data is far smaller than that of the original data, compression of the hyperspectral data is thus realized, and the spatial structure characteristics and the spectral structure characteristics of the hyperspectral data are made full use of.

The invention relates to a hyperspectral image compressed coding method, in particular a hyperspectral image compressed coding method through multivariate vector quantization. The hyperspectral image compressed coding method through the multivariate vector quantization comprises the steps of reading hyperspectral image data; constructing a compression dictionary: according to two multivariate vector quantitative models and dictionary atom selection strategies, calculating dictionary atoms and coefficients of the dictionary atoms used by compression of each of picture elements by means of multiple regression; performing compressing and coding. According to the method, the vector quantitative models are reconstructed, two multivariate vector quantitative models are provided, the omitted error information is less than that in the traditional vector quantitative method, coefficients of the selected dictionary atoms are not defined in the newly-provided multivariate vector quantitative models, values of the coefficients are calculated out according to a spectrum and the dictionary, therefore, quality of the reconstructed images is ensured, and image distortion caused by compressed coding is reduced.

The invention provides a hyperspectral image compression method and application. The method comprises the following steps: (1) extracting a spectral curve of the target zone pixel of a hyperspectral image; (2) carrying out discrete Fourier transform to the spectral curve, and decomposing the spectral curve into a plurality of linear-combination trigonometric function waves; and (3) selecting the first p trigonometric function waves with the maximum amplitude, which corresponds to each pixel, extracting the amplitudes, the frequencies and the initial phases of the p trigonometric function waves, and combining to serve as a characteristic value of the pixel, wherein p is a natural number less than or equal to 5. Through the Fourier transform, the characteristic values of each pixel are reduced to 3*p, the original characteristic value length and the waveband number of each pixel are the same, a compression ratio is high, and precision regulation can be realized by changing the value of p. The invention is used for object classification and has the advantages of being simple in data handling capacity, easy in implementation and high in identification precision.

The invention discloses a high spectral image compression sensing method based on manifold structuring sparse prior and solves a technical problem of low precision existing in a high spectral image compression sensing method in the prior art. The method is characterized in that a few linear observation values of each pixel spectrum are sampled randomly and are taken as compression data, through the manifold structuring sparse prior, sparsity of a high spectral image after sparsification in the spectrum dimension and manifold structure of the high spectral image in the space dimension are etched, through a hidden variable Bayes model, signal reconstruction is carried out, and sparse prior learning and noise estimation are unified to one regularization regression model for optimization solution. The sparse prior acquired through learning can not only fully describe the three-dimensional structure of the high spectral image, but also has relatively strong noise robustness. The sparse prior is utilized to realize high precision reconstruction of the high spectral image. Based on tests, Gauss white noise is added to the compression data to make the signal to noise ratio of the compression data to be 15db, the sampling rate is 0.09, and thereby the 23db peak value signal to noise ratio is acquired.

The invention discloses a hyperspectral image compression and classification method based on discriminative feature learning, which is used for solving the technical problem of poor practicability ofthe existing hyperspectral image compression and classification method. According to the technical scheme, the end-to-end compressed classification network comprises two branch structures, wherein onepath is a stack self-encoding module and is used for learning data identifiability features, the encoder is used for feature compression, and the decoder is used for feature decompression. A mean square error loss function of all data is calculated through an encoder and a decoder; and the other path is a classification module which is used for classifying identifiable features, the encoder and the classifier use a shared module, and the classifier performs classification by utilizing identifiable compression features obtained by the encoder so as to complete end-to-end feature compression and classification tasks. The shared encoder module not only can obtain identifiable characteristics, but also can efficiently perform a hyperspectral image classification task according to the identifiable characteristics, and is good in practicability.

The invention discloses a hyperspectral image compression method based on interested pixel protection. The method comprises two steps, wherein the first step refers to interested pixel extraction, and the second step refers to compression method implementation based on interested pixel reservation. By the adoption of a deblurring method, main edges of a hyperspectral image are protected. The hyperspectral image compression method based on the interested pixel protection is superior to a common compression method, and effectiveness of the hyperspectral image compression method is fully proved through experimental results. In other regions, edge information acquired through the hyperspectral image compression method is more detailed than information obtained through the common compression method, and the advantages of the hyperspectral image compression method are displayed more remarkably as the compression ratio increases in spite of differences in degree.

The invention discloses a full-polarization hyperspectral image compression and reconstruction method based on Stokes parameter partitioning, and the method can increase the degree of freedom of polarization compression, improve the pertinence and reconstruction precision of polarization reconstruction and shorten the overall reconstruction time. A to-be-detected full-polarization hyperspectral image is imaged on a detector by adopting the combination of a quarter-wave plate and a liquid crystal adjustable filter, and different full-polarization modulation modes are realized by selecting the fast axis angle of the quarter-wave plate and the incident plane angle of the liquid crystal adjustable filter. For the combination of one fast axis angle and two incident plane angles, a first Stokesparameter S0 is solved by using a summation method, and the last three Stokes parameters S1S2S3 are reconstructed; for the combination of 2-3 fast axis angles and 1 incident plane angle, S1S2S3 are reconstructed by using a differencing method, and then S0 is solved; and for the combination of 1-3 fast axis angles and 1 incident plane angle, S0 is reconstructed by using a scaling method, and then S1S2S3 are reconstructed. And finally, a reconstructed full-polarization hyperspectral image is obtained.

The invention provides an image high-speed compression method and system based on an FPGA (Field Programmable Gate Array) under a CCSDS (Consultative Committee for Space Data System) standard. Theimage high-speed compression method comprises the following steps of: firstly, extracting pixel values in spectral image data, generating a region mark signal and a spectral band mark signal of each pixel value, predicting a weight vector of each pixel according to a comparison mark bit and a predicted value of a central local difference by adopting a forward prediction mode, and obtaining an error value between a pixel true value and a predicted value, and then coding all error values in a Golomb-Rice coding mode to realize lossless compression of image data. According to the method and system, a lossless data compression algorithm of the CCSDS120.2-G-1 standard is analyzed, a forward prediction mode is adopted according to the characteristics of a feedback loop of the algorithm, meanwhile, the calculated amount in a key path is reduced, a full-pipeline structure of hardware is achieved; the method and system can be applied to hyperspectral image compression in the fields of geological exploration, agricultural research, satellite-borne remote sensing image research and the like, and is particularly suitable for real-time compression of hyperspectral image data of satellites or space stations.