220 results about "Spectral dimension" patented technology

The invention provides a spectral value calculation model training method and device, a medium and equipment, and the method comprises the steps: determining a target spectral dimension, obtaining a first structure parameter set as a training sample, and taking a corresponding spectral value as an annotation; and constructing a to-be-trained spectrum value calculation model according to the target spectrum dimension, inputting a training sample into the spectrum value calculation model, and training the spectrum value calculation model by taking the minimum difference between output and annotation as an optimization target. And according to the determined target dimension number, constructing the to-be-trained spectrum value calculation model, so that the trained spectrum value calculation model outputs the spectrum values of the target dimension number, and the efficiency of constructing the native database is improved. In addition, the size of the spectral value calculation model is far smaller than that of the native database, so that when optical critical dimension measurement is carried out on different devices, the spectral value calculation model can be migrated in the different devices to replace migration of the native database, the migration speed is increased, and storage resources of the devices are saved.

The invention relates to a multispectral image and hyperspectral image fusion method and system, and belongs to the field of deep learning and signal processing, and the method comprises the steps: carrying out the coding and feature extraction of a multispectral image and a hyperspectral image through employing a coder based on a spectrum self-attention mechanism; the hyperspectral image features and the multispectral image features obtained by the encoder are input into a space-spectrum fusion module, deep fusion of the spatial dimension and the spectral dimension is carried out, and unified feature representation is obtained; inputting unified feature representation obtained by fusion into a decoder based on a spectrum self-attention mechanism, inputting layered features passing through a graph-based spectrum sensing module into a corresponding decoder symmetric layer, and then decoding and outputting to obtain a spectrum image with high spatial resolution and a spectrum image with high spectral resolution; through the above strategy, full fusion and complementation of spectral image space information and spectral information are ensured, and the method has important value for research and application of MHIF.

The invention relates to the technical field of power distribution station inspection, and discloses a power distribution station automatic inspection method based on image recognition. According to the method, multi-spectral image data of multiple areas in a power distribution station are collected in real time, and a multi-channel feature tensor of an equipment state is generated through a feature extraction network; performing feature fusion of space and frequency spectrum dimensions on the multichannel feature tensor by using a multi-scale convolutional attention network, and outputting an enhanced device feature map; inputting the data into a cascade anomaly detection module, positioning an equipment surface defect region by adopting a region segmentation algorithm, and analyzing and generating a defect evolution trend vector in combination with time sequence characteristics; on the basis of the vector, probability distribution of equipment fault risks is predicted through a space-time propagation model, and a dynamic risk field is generated; and finally, constructing a self-adaptive early warning decision tree for the dynamic risk field, and generating an inspection maintenance instruction according to risk probability threshold grading. According to the method, automation and intelligentization of power distribution station inspection are realized, and support is provided for efficient maintenance of the power distribution station.

The invention provides a multi-source remote sensing data classification method, which belongs to the field of hyperspectral image processing in the field of remote sensing, and comprises the following steps: acquiring remote sensing data of various target objects; extracting shallow layer features of each kind of remote sensing data; obtaining a plurality of frequency characteristics of each kind of remote sensing data; fusing the same frequency features of all remote sensing data to obtain a plurality of same frequency fusion features; splicing the plurality of same-frequency fusion features to obtain a multi-source fusion feature; enabling the multi-source fusion features to sequentially pass through a frequency modulation layer and an attention layer which are superposed to obtain fused global features and local features; and weighting the global features and the local features in the spectral dimension to obtain a prediction classification result of the target object. According to the method, directional frequency characteristic decomposition and fusion of the multi-source data are realized, the multi-source fusion characteristics are extracted based on the multi-source remote sensing data to serve as the basis of classification, more comprehensive information is provided for classification of target objects, complex ground object characteristics can be fully captured, and the classification result is more accurate.

The invention provides a hyperspectral spatial-spectral fusion classification method, a hyperspectral fast detection device and a storage medium, and the method comprises the steps: employing a twin network layer, and enabling two convolutional neural network branches of the twin network layer to extract the feature value of the spatial dimension and the feature value of the spectral dimension of hyperspectral data respectively, the two convolutional neural network branches share a weight; inputting the feature value of the spatial dimension and the feature value of the spectral dimension into a feature fusion layer for fusion to form a fusion feature vector; inputting the fusion feature vector into a shared network layer, and carrying out global modeling by using a Transform model; and inputting the output features of the Transform model into a classifier formed by a full connection layer, and carrying out final classification prediction. According to the method, the spatial-spectral characteristics of the hyperspectral data are fully utilized, and the classification precision and robustness are improved.

The invention discloses an ultra-high-speed imaging device based on acousto-optic filtering modulation. The ultra-high-speed imaging device comprises a spectrum modulation system, a spectrum imaging system and a data processing system. The spectrum modulation system utilizes an acousto-optic tunable filter (AOTF) to dynamically modulate broadband continuous spectrum light, narrow-band illumination light which rapidly changes along with time is generated, and time information of a dynamic scene is effectively mapped to a spectrum dimension. And the spectral imaging system images the spectral coded dynamic scene to the hyperspectral camera to realize acquisition of spectral space-time information. The data processing system performs spectral crosstalk correction, channel separation and demosaicing reconstruction on the acquired original mosaic image, and recovers a time sequence image according to a spectrum-time mapping relation, thereby realizing single-shot superspeed imaging in a dynamic process. The device has the advantages of being simple and compact in structure, flexible in time window, high in imaging fidelity and the like, and has wide application prospects in the fields of rapid dynamics such as microfluidics, laser processing and ultrafast physics.

The invention relates to the technical field of image recognition, in particular to a red tide algae recognition method based on a hyperspectral image, and the method comprises the following steps: calculating a gradient value of each pixel in a spatial dimension and a spectral dimension according to an original hyperspectral data cube, setting a diffusion coefficient for diffusion, and obtaining a three-dimensional fidelity hyperspectral data cube. According to the method, the gradient change trends of the spatial dimension and the spectral dimension are extracted at the same time pixel by pixel, and the regional differentiation diffusion operation is matched, so that the expression ability of the alga plaque boundary and the consistency of the internal spectral characteristics are enhanced, and the fidelity of the hyperspectral data in fine-grained region identification is improved; upper convex hull calculation of a spectral reflectivity curve and construction of a continuous spectrum background baseline are introduced, and characterization of pigment absorption morphological characteristics is enhanced; and in combination with the characteristic absorption wave bands of chlorophyll and phycobiliary, the synergistic weighted response to the distribution of the algae pigment is realized, and the sensitivity and selectivity of pigment recognition are improved.

The invention provides a hyperspectral image reconstruction method and device based on high compression ratio snapshot compression and a readable storage medium thereof, and an end-to-end reconstruction method based on a lightweight space-spectrum aggregation block (SSAB). The SSAB includes a packet hybrid space module (GHSB), a bi-directional patch spectrum module (BPSB), and a gated feedforward network (GFFN). The GHSB performs multi-direction scanning along channel grouping, compensates information isolation through residual connection, and efficiently models spatial dependence; the BPSB divides the features into blocks and then splices the blocks along the spectral dimension, and spectral context dependence is captured through bidirectional scanning; the GFFN enhances cross-band nonlinear interaction through a gating mechanism, and optimizes space-spectrum feature fusion. According to the method, high-precision hyperspectral image reconstruction is realized in a high-compression-ratio scene, the calculation complexity and the parameter scale are reduced, the reconstruction efficiency and quality are effectively improved, and the method is suitable for unmanned aerial vehicle remote sensing, real-time monitoring and other scenes with limitation on calculation resources.

The invention discloses a scene-based push-broom hyperspectral image adaptive non-uniformity correction method, which comprises the following steps of: automatically sorting each scanning line of a wave band of a hyperspectral image from small to large according to pixel DN values of different spatial dimensions, and adaptively selecting at least one dark uniform region and one bright uniform region of the spatial dimensions; a correction coefficient of each spatial dimension pixel is obtained, and non-uniform correction of each spatial dimension is completed; on the basis of spectral dimension characteristics, spectral dimension non-uniform correction is completed according to the spatial dimension non-uniform correction result, spatial dimension non-uniform correction is repeated according to the spectral dimension non-uniform correction result, and hyperspectral image non-uniform correction is completed. The method is applied in a full-automatic mode, limitation of scenes can be broken through, space and spectrum combined information is used for non-uniform correction, space structure information and spectrum information of the hyperspectral images are reserved to the maximum extent, the quality of the hyperspectral images can be effectively improved, and the method is suitable for self-adaptive non-uniform correction of mass hyperspectral images of various complex scenes. And the method has good universality.

The invention relates to the technical field of hyperspectral image processing, in particular to a hyperspectral image classification model. The objective of the invention is to solve the problems of computation complexity and performance bottleneck in hyperspectral image processing. The model can be applied to hyperspectral image analysis tasks in the fields of remote sensing, environment monitoring, resource exploration and the like. Efficient dimension reduction processing is carried out by adopting random Fourier feature nonlinear principal component analysis (RFF-NLPCA), redundant information is effectively reduced, and the calculation speed is increased. A lightweight space SSM Block module and a spectrum SSM Block module are introduced into the model, and the image classification precision and robustness are improved through feature modeling of the space dimension and the spectrum dimension. The gating fusion module (GFM) optimizes the fusion capability of space and spectral features, and improves the discrimination capability of feature expression. Finally, the model can efficiently process high-dimensional hyperspectral data, provides a more accurate classification result, has relatively low calculation overhead and relatively high performance, and is particularly suitable for processing large-scale hyperspectral image data.

The invention discloses a hyperspectral image particle segmentation method and device, relates to the field of image processing, and is used for improving the accuracy of particle segmentation according to a hyperspectral image. The method comprises the steps that a local area corresponding to pixels is selected, a preset matrix of the pixels is constructed, the preset matrix is formed by covariance of any two spectral vectors in spectral vectors of the pixels in the local area corresponding to the pixels, and whether the pixels are candidate edge pixels or not is judged according to the characteristic value condition of the preset matrix; compared with the existing edge detection method which only utilizes the spatial gradient information of the image, the method has the advantages that the accuracy of edge pixel detection can be improved by judging whether the pixel is the candidate edge pixel or not through combining the characteristics of the spatial dimension and the spectral dimension; and the particle region is further segmented according to the candidate edge pixels, so that the edge pixels are detected by combining the features of the spatial dimension and the spectral dimension and the particle region is segmented according to the edge pixels, and the accuracy of performing particle segmentation according to the hyperspectral image can be improved.

The invention discloses a hyperspectral image cross-scene classification method based on dynamic mark enhanced mamba, comprising the following steps: inputting an obtained target domain image into a trained classification model, and outputting a prediction category of the target domain image; wherein the training of the classification model comprises the following steps of: 1, expanding source domain image samples into block sequence data in a spatial dimension in sequence, and adding a position code and a spectral dimension projection through a block embedding layer; 2, performing dynamic mark enhancement on the features of the data obtained after projection; step 3, extracting the features after dynamic mark enhancement; 4, repeating the step 2 and the step 3 for N times; 5, classifying the finally extracted features through a classifier head, obtaining the prediction probability of a sample corresponding to each feature, calculating classification compensation loss, and adjusting classifier parameters by using the classification compensation loss as a feedback signal; the method has the advantages of being good in generalization ability, high in classification precision and high in efficiency.

The invention discloses a small sample HSI classification method based on prototype mask comparison and cross-stage feature refinement, and the method comprises the steps: carrying out the meta-task training in a source domain and a target domain in a crossing manner, carrying out the spectral dimension unification of samples of a support set and a query set in each meta-task, carrying out the multi-dimensional feature fusion resistant to the dependence of adjacent pixels, and carrying out the classification of the multi-dimensional features. Obtaining support features and query features; constructing positive and negative sample pairs representing a prototype under each meta-task, and performing prototype mask comparison through graph convolution; according to the method, feature redundancy elimination refining is carried out based on emphasis mapping of a cross-domain shared multi-layer perceptron, cross-stage feature refining is carried out on support features and query features through the multi-layer perceptron and directional calibration coordination, refined support features and refined query features are obtained, and different losses are optimized in two training stages; in the test stage, small samples of a target domain are used, and a K-nearest neighbor classifier is trained for classification prediction through features extracted after dimensionality unified mapping, a feature extraction module and a multi-layer perceptron in sequence.

The invention relates to a spatial spectrum dimension scanning method, in particular to a Mamba spatial spectrum dimension scanning method for a hyperspectral image classification task, which comprises the following steps of: 1, for an input hyperspectral image tensor, performing rotation and turnover construction on an input image through four spatial directions to obtain a hyperspectral image tensor; step 2, performing interval snakelike scanning on the view in each direction; performing interval snakelike scanning: reserving pixel points in even columns in even rows; pixel points of odd columns are reserved in odd rows, and the sequence of the selected pixels in the row is reversed so as to simulate the scanning characteristic of the snakelike path. And the feature expression capability and the calculation efficiency in a hyperspectral image classification task are improved. A scanning mechanism of jump sampling and snakelike traversal is introduced in the spatial dimension, redundant sampling in the spatial dimension can be effectively reduced, spatial structure information of the image is reserved, and the classification precision of the hyperspectral image is improved on the premise of ensuring the integrity of the image spectrum and the spatial information.

In remote sensing, hyperspectral (HS) images are acquired to study earth's surface. Processing of HS images through neural networks demands significant computational resources for both training and inference phases. The present disclosure addresses the unresolved problem of the conventional methods for reducing dimensions in hyper-spectral data and accelerating the training and inference of a model by applying approximate computing techniques. The approximate computing techniques leverage physical properties of a reflectance spectra. This makes the HS images interpretable across various applications. In the present disclosure, three reflexivity-based approximate computing techniques namely R-Hop(K), R-Top(N), and R-Proximity(N) are implemented. These reflexivity-based approximate computing techniques use spectral clustering methods that rely on reflectance values to capture inherent characteristics from hyperspectral images across diverse domains. Further, existing spatial dimension reduction techniques and a combination of spatial and spectral dimension reduction techniques are evaluated.

The invention provides an intelligent cashmere doping content judgment system utilizing multi-mode Raman spectrum characteristics, and relates to the technical field of intelligent spectrum analysis. The cashmere doping content intelligent judgment system utilizing the multi-mode Raman spectrum characteristics comprises a data acquisition module, a multi-wavelength laser excitation unit is used for carrying out space scanning on a sample, time sequence Raman spectrums under different polarization configurations are synchronously acquired, and the time sequence Raman spectrums are analyzed; multi-dimensional original data including space coordinates, a time sequence, spectral intensity and a polarization state is formed, and an original multi-modal data set is generated. Through multi-wavelength laser space scanning and polarization time sequence synchronous acquisition, a multi-dimensional original data set containing space coordinates, a time sequence, spectral intensity and a polarization state is constructed, the dimension limitation of a traditional spectrum is broken through, and a three-dimensional convolution kernel is adopted to perform synchronous sliding processing in space and spectrum dimensions, so that the spectrum intensity and the polarization state are obtained. And the spatial distribution rule and multi-mode spectrum correlation characteristics of the doped region are directly captured.

The invention relates to a multi-source remote sensing cooperative system based on an unmanned aerial vehicle formation and a control method, and the system can synchronously collect laser radar point cloud data, multispectral images, thermal infrared images and hyperspectral images, and breaks through the limitation of single data dimension caused by the fact that a traditional single unmanned aerial vehicle platform can only carry a single sensor. Meanwhile, the formation control module controls the heterogeneous unmanned aerial vehicle cluster to form a preset formation and perform cooperative flight to ensure spatial position cooperation of the unmanned aerial vehicles in the operation process, and the task allocation module dynamically allocates route tasks and sensor working parameters to the unmanned aerial vehicles based on target area environment information. And the data fusion module performs geometric registration, spectrum fusion and three-dimensional reconstruction processing on the multi-source remote sensing data acquired by each unmanned aerial vehicle to generate a three-dimensional spectrogram of the target area, so that accurate integration of the multi-source data in space and spectrum dimensions is realized, and the multi-source remote sensing data is acquired. Therefore, the system gives consideration to large-range area coverage capability and high monitoring precision.

The invention discloses a hyperspectral image and laser radar data fusion classification method based on an improved attention mechanism. The method comprises the following steps: obtaining hyperspectral image data and laser radar elevation data to be classified; processing the hyperspectral image data through a spectral channel attention module to extract spectral features; processing the laser radar elevation data through an elevation space attention module to extract spatial features; inputting the spectral features and the spatial features into a double-feature fusion module for coupling to generate fusion features; image classification is completed based on the fusion features; according to the method, the image features of the hyperspectral image in the spectral dimension and the elevation information of the radar image in the spatial dimension can be fully extracted, and the representation capability of the model for the spatial features and the spectral features is enhanced.

The invention discloses a radiation calibration method and system for a spectral imaging device, and the method comprises the steps: constructing a radiation response model of a pixel of the spectral imaging device, enabling target light to change in a spectral dimension and a radiation dimension, obtaining the spectral radiance data of the target light and the response data of the pixel of the spectral imaging device under the target light with different spectrums or / and radiance, and obtaining the spectral radiance data of the target light and the response data of the pixel of the spectral imaging device. According to the spectral radiance data, obtaining an equivalent radiance value of the multi-band light received by the pixel, and then according to the equivalent radiance value of the multi-band light received by the pixel and response data of the pixel, obtaining a radiation response coefficient value; and obtaining a radiation response model for describing the relation between the response value of the pixel and the radiance of the light of multiple wavebands received by the pixel. The radiation calibration of the spectral imaging device for coupling the spectral information and the spatial information is realized.

The invention discloses a computing spectrum reconstruction method based on CS-Unet and a spectral imaging system. The calculation spectrum reconstruction method comprises two parts, namely TwIST pre-reconstruction and Unet coupling reconstruction, wherein the TwIST pre-reconstruction is used for preprocessing an initial coding spectrum image to generate a pre-reconstructed spectrum image; and the Unet part couples the pre-reconstructed image and the initial coding spectral image in the spectral dimension and inputs the coupled image and the initial coding spectral image into a network for fine reconstruction, and finally a target space-spectral image is recovered. The imaging system is composed of a spectral image coding module and a CS-Unet reconstruction module, a metasurface spectral modulator applied to the spectral image coding module is composed of 36 silicon-based all-dielectric structures, and rich spectral response is achieved by adjusting the period, the size and the direction angle of three basic C4 symmetric units, namely a round unit, a cross unit and a square unit. According to the method, high-robustness and high-imaging-quality spectrum reconstruction is realized in a visible light wave band (400-700 nm), and the method has relatively high engineering applicability.

The invention relates to a sea surface small target floating object detection method based on an RGB reconstruction hyperspectrum, and the method comprises the steps: obtaining an RGB image, carrying out the preprocessing of the obtained image, and constructing an RGB image sample data set through the preprocessed image; generating a hyperspectral data set of a corresponding scene from the RGB image sample data set by using the residual dense network integrated with the frequency domain information; 31 wave bands of the hyperspectral image are grouped, and the image with the highest entropy value in each group is selected to synthesize a pseudo-color image; and inputting the pseudo-color image into the fixed detection network to obtain a target detection result. According to the invention, the detection of the small target floating object on the sea surface is realized by using the spectral dimension information of the hyperspectral image.

The invention provides a hyperspectral tailing pond dry beach line intelligent analysis method, and belongs to the technical field of image processing. Through multiband characteristics of a hyperspectral remote sensing image, principal component clustering analysis and characteristic wavelength selection are combined, so that the problems of insufficient spectral information of the image after dimension reduction and serious data redundancy caused by use of a full-band image are solved; single-band images corresponding to different characteristic bands emphasize different target information in the original image, and different combinations are beneficial to fully exerting the advantages of different target information of each band, so that a water area and a dry beach are distinguished more accurately; a double-branch network model is adopted, the difference between spectral dimensions and spatial dimensions is synchronously analyzed, spectral features and spatial information are fused, and the two branches have certain complementarity, so that rich spatial spectrum information of an input hyperspectral remote sensing image can be fully excavated, a dry beach line can be identified more accurately and effectively, the length of the dry beach is calculated, manual intervention is reduced, and the working efficiency is improved. The treatment efficiency is improved.

Optically computed phase imaging systems and methods are provided. An example system includes an interferometer configured to output 3D spatial-spectral data of a sample including an interferometric signal and an optical computation assembly having a spatial light modulator configured to modulate the 3D spatial-spectral data with a modulation pattern. The modulation pattern modulates the interferometric signal along a spectral dimension to select a depth to obtain a sample signal at the selected depth and modulates the interferometric signal along a first spatial dimension to create a waveform to facilitate phase extraction. The system further includes a detector configured to detect 2D spatial data of the sample. The system further includes a processor coupled to a memory, the processor configured to process the 2D spatial data to extract phase information of the sample.

The invention relates to a hyperspectral image super-resolution method and system based on spectral space decoupling attention, and aims to solve the contradictory problem of insufficient space-spectrum cross-dimension feature collaborative modeling and serious deep network calculation redundancy in the existing single-image super-resolution technology. According to the method, a double-stage decoupling attention mechanism and a feature multiplexing architecture are innovatively designed by constructing a superficial layer space spectrum unwrapping attention network (SSDAN): firstly, a space-spectrum decoupling attention (SSDA) module composed of cascaded banded space attention and grouped spectrum attention is adopted; through spatial dimension direction perception feature extraction and spectral dimension long-range correlation modeling, decoupling type cooperative expression of global spatial features and cross-band spectral features is realized; and secondly, constructing a cross-layer feature dynamic aggregation mechanism, and realizing high-precision reconstruction under an extremely shallow architecture only comprising two SSDA layers by strategically reusing intermediate features of a shallow network.

The invention discloses a multi-spectral target recognition method based on double-branch multi-view spatial-spectral feature interaction, which comprises the following steps: performing principal component analysis operation on a multi-spectral image to perform spectral-spectral dimension reduction processing, and extracting a three-dimensional image block as input data by using pixel-level segmentation operation; capturing shallow spectrum-space joint features of the input data; extracting local and global spectrum-space joint features of the input data by using a CNN (Convolutional Neural Network); extracting local and global correlation characteristics between a spectrum band and a space pixel by utilizing Transform; the self-adaptive interaction of local and global spectrum-space joint features captured by CNN and Transform is fused in an early stage, and interdependent attention with more discriminant characteristics is generated; and an output module is utilized to obtain a multi-spectrum target recognition result through mutual dependence on attention.

The invention relates to a Raman denoising method. The method comprises the following steps: encoding an initial feature map of a Raman spectrum or a Raman image to obtain an encoded feature map; wherein the coding processing comprises the steps of performing feature extraction processing on the initial feature map to obtain a coding separation feature map, and performing feature fusion processing on the coding separation feature map to obtain a coding feature map; carrying out dimension integration and feature integration processing on the coding feature map to obtain an undertaking feature map; decoding the undertaking feature map to obtain a decoded feature map after noise of the corresponding Raman spectrum or Raman image is filtered; wherein the decoding processing comprises the steps of carrying out expansion processing on the coding feature map to obtain a sampling feature map, splicing the sampling feature map and the coding separation feature map to obtain a decoding splicing feature map, and carrying out feature extraction processing on the decoding splicing feature map to obtain a decoding feature map. According to the invention, the Raman spectrum or the Raman image can be denoised in consideration of the spatial dimension and the spectral dimension, and the denoising effect is strong.

The present invention discloses a spectral smoothing method applicable to infrared spectral data. The steps of the method are as follows: (1) Determine whether the infrared spectral data is sampled at equal intervals in the spectral dimension; (2) If the infrared spectral data is non-uniformly sampled, perform uniform resampling on the infrared spectral data; (3) Perform mean filtering on the uniformly sampled infrared spectral data; (4) Calculate the spectral similarity before and after mean filtering; (5) Determine whether the spectral similarity is within the threshold range. If it is not within the threshold range, continue to perform mean filtering; if it is within the threshold range, output the smoothed infrared spectral data. The method of the present invention has the characteristics of good accuracy, easy operation, and easy determination of parameters.

The present invention relates to a hyperspectral image classification method based on a spectral-spatial attention mechanism of a dual-branch network, including: inputting hyperspectral image data cubes; using principal component analysis to reduce the spectral dimension of the hyperspectral image, and for the pixels to be classified, encapsulating them into adjacent region blocks as the input of the spectral sub-network; creating an adjacent region around each specific pixel to collect spatial information and using it as the input of the spatial sub-network; obtaining one-dimensional spectral features; obtaining one-dimensional spatial features; fusing and balancing the one-dimensional spectral features and the one-dimensional spatial features through a fusion layer, and using a softmax regression layer to predict the probability distribution of each type of ground object. By taking the hyperspectral image as the research object, the present invention improves the effective feature representation of the hyperspectral image under the premise of limited samples, ensuring both the classification accuracy and reducing the number of parameters and the computational amount of the network model; fully utilizing the spectral information and the spatial information to obtain better classification results.

The application discloses a hyperspectral image Transformer network training and classification method, which first divides a hyperspectral image sample based on a spectral dimension to obtain a plurality of spectral subbands, wherein the hyperspectral image sample is an unlabeled training sample; each spectral subband is input into an embedding module to obtain local space-spectrum embedding features output by each embedding module, wherein the embedding module extracts space-spectrum features of the spectral subband at multiple scales; all local space-spectrum embedding features are fused according to the positions of the spectral subbands to obtain global space-spectrum embedding features; the global space-spectrum embedding features are input into a Transformer encoder, and a center region token is masked and reconstructed in a Transformer decoder to train the Transformer encoder in a self-supervised manner. Compared with the prior art, the unlabeled sample can be effectively utilized to improve the training effect of the Transformer network, and the hyperspectral image can be classified with high precision.

The invention provides a light-splitting imaging system, and relates to the technical field of optics, the light-splitting imaging system comprises a light-splitting module and a detection device, the light-splitting module comprises a light path conduction assembly and a plurality of light-splitting modules, and the plurality of light-splitting modules are distributed on a propagation path of a light signal; the detection device is arranged on the light emitting sides of the light splitting modules and comprises a plurality of detection units which are in one-to-one correspondence with the light splitting signals. According to the invention, a plurality of light splitting modules can be adopted to separate the light signal into a plurality of light splitting signals according to different wavelengths and / or polarizations, each light splitting signal can keep the whole image and carry wavelength or polarization information at the same time, and a light signal array of a spectral dimension is expanded to a two-dimensional space. Therefore, the optical information of the spectral dimension can be obtained through one-time shooting of a plurality of detection units, image splicing processing is not needed, and the acquisition speed of the system can be greatly improved.