17 results about "Spectral dimension" patented technology

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.

This invention relates to the field of optoelectronic detection technology and discloses an online detection system for screen color uniformity based on dynamic rotating polarization spectrum. The system includes an online transmission module for carrying and driving the illuminated screen under test to move continuously along a set direction at a set speed; and a combined optical encoding module, positioned above the screen under test, comprising a static spatial-spectral phase delay mask and a continuously rotating polarization modulator arranged sequentially along the beam propagation direction. By driving the screen to continuously translate using the online transmission module, and cooperating with the continuously rotating polarization modulator and hyperspectral acquisition module to acquire dynamic spatiotemporal aliasing data, the accuracy of the spectral dimension can be guaranteed. Furthermore, by extracting high-precision chromaticity parameter matrices and multidimensional polarization distortion feature matrices from the decoupled and reconstructed optimal polarization spectral tensor solution, the system's ability to detect complex internal quality problems can be significantly enhanced.

This invention discloses a method for classifying defects in patternless wafers based on multi-feature joint decision-making, belonging to the field of wafer defect detection technology. The method includes the following steps: irradiating the surface of a patternless wafer with a probe beam having a preset polarization state; acquiring spatial scattered light intensity information of the scattered light from the wafer surface under different combinations of polarization states based on a probe array distributed at multiple spatial angles; extracting local optical features of the candidate defect regions and constructing a feature tensor based on these local optical features; analyzing the candidate defect regions according to the polarization degradation evolution law of the feature tensor in the spectral dimension and outputting the defect classification result. This approach avoids the low classification accuracy caused by highly similar scattering features of small defects, significantly improving the accuracy of the detection results.

The application belongs to the technical field of remote sensing data processing, and discloses a hyperspectral and laser radar data joint ground feature classification method, equipment and medium, the method comprises the following steps: acquiring hyperspectral image data and laser radar elevation data; feature extraction is carried out by adopting a three-branch parallel structure, wherein the hyperspectral image branch extracts global spectral spatial features by three-dimensional convolution and bidirectional scanning along the spectral dimension, the laser radar branch extracts elevation structure features by multi-level convolution, and the multi-modal branch extracts multi-modal fusion features by early fusion and a multi-scale feature enhancement module; a multi-head bidirectional cross attention module is used to perform deep bidirectional interaction and fusion on the features extracted by the three branches, thereby generating a final fusion feature vector; finally, the final fusion feature vector is input into a classifier to obtain a classification result. The method realizes deep interaction between modes and adaptive perception of multi-scale ground features, and significantly improves the accuracy of joint classification.

This invention relates to the fields of aerospace remote sensing and satellite overall design technology, and particularly to a method for analyzing spectral aliasing caused by flywheel micro-vibration in hyperspectral imagers. The method involves acquiring flywheel operating parameters and satellite structural dynamics parameters, calculating the time series of line-of-sight jitter displacement caused by flywheel micro-vibration, acquiring a standard hyperspectral data cube of the target scene, performing a dynamic spectral sampling process using the line-of-sight jitter displacement time series, continuously integrating the data cube by simulating the instantaneous line-of-sight motion trajectory of the imager, and outputting a hyperspectral image affected by micro-vibration. The method also involves extracting the spectral curves of the pixels to be evaluated and the ideal spectral curve, and calculating the spectral aliasing error index. This invention quantifies the impact of micro-vibration on spectral dimensions by constructing a complete analytical chain of flywheel disturbance, structural transmission, and dynamic sampling, solving the problem of the lack of quantitative analysis of spectral aliasing in existing technologies, and significantly improving the reliability of hyperspectral remote sensing data in agricultural crop classification and growth monitoring.

The application discloses a kind of ocean surface cloud detection methods based on quantum convolutional neural network, it is related to cloud detection technical field, and each channel reflectivity value is handled as the angle input of rotary door in encoding layer after normalization mapping;Two features are encoded to each quantum bit by Rx and Ry door operation;Adjacent quantum bits realize spatial feature association by CZ door, simulate the local connection characteristics of classic convolution;Quantum bits are operated by pooling;The state of the two quantum bits left after multiple measurement convolution and pooling operation, obtain the measurement expectation value of the two quantum bits, carry out cloud detection task by training quantum convolutional neural network, with better generalization ability and less parameter quantity;Driven by quantum parallelism, single quantum convolution operation can act on the entire spectral dimension simultaneously, avoiding the redundant steps of serial computation in classic CNN per channel.

This invention provides a method and system for intelligent human-computer interaction interpretation of landslide ridges with 3D visualization and orthophoto synchronization. It constructs and dynamically links a high-dimensional, multi-feature space integrating planar coordinates and elevation, including spectral dimensions across multiple bands (blue, green, red, and near-infrared), and incorporating the temporal dimension and derived feature dimensions of multi-temporal remote sensing observation data. This space serves as the data foundation and driving engine for the entire interpretation process. Through bidirectional synchronous view linkage, feature linkage, human-computer intelligent linkage, and judgment process linkage between 2D orthophoto views and 3D real terrain scenes, it achieves accurate extraction of landslide ridge boundary vectorization. Based on this, the system can intelligently identify and present landslide ridge morphological features, accurately determine the development status of landslide ridges, and automatically mark their risk level upon completion of the landslide determination. This invention significantly improves interpretation accuracy and work efficiency while providing an intuitive and user-friendly human-computer interaction experience.

The present application relates to hyperspectral remote sensing image processing technical field, specifically to a kind of hyperspectral remote sensing image recognition method, system and medium based on superpixel, wherein hyperspectral remote sensing image recognition method includes the following steps: 1, the source hyperspectral image is carried out spectral dimension reduction, obtain the hyperspectral image after dimension reduction;2, extract superpixel information from source hyperspectral image, and obtain superpixel level class probability, then the superpixel level class probability is spatially optimized, and the optimization probability of different ground objects is obtained;3, extract structural features from the hyperspectral image after dimension reduction, and obtain feature level class probability;4, the optimization probability of different ground objects and feature level class probability are fused, and the final classification result is obtained.The present application can fully and accurately mine spatial information in superpixel, and fuse edge information to optimize ground object misclassification problem, and simultaneously, the present application can obtain higher ground object classification precision and better visual effect.

This specification provides a method for evaluating and training a spectral computation model, as well as a method for measuring structural parameters. The method includes: acquiring a training dataset, wherein each training sample in the training dataset includes a set of structural parameters describing a grating structure and a first theoretical spectrum of a first spectral dimension; inputting the structural parameters into a spectral computation model to output a predicted spectrum of a target spectral dimension; calculating a loss value between the predicted spectrum and a second theoretical spectrum of the target spectral dimension, wherein the second theoretical spectrum is obtained by dimensionality reduction of the first theoretical spectrum, and the loss value is used to train the spectral computation model; and multiplying the loss value by a correction factor to obtain a correction evaluation index, which is used to evaluate the performance of the spectral computation model, wherein the correction factor is the ratio of the target spectral dimension to the first spectral dimension.

The present application relates to the field of image classification, and particularly relates to a hyperspectral image classification method, comprising: performing joint dimension reduction processing on hyperspectral image data by using FPCA and a three-dimensional convolution kernel of 1*1*d to obtain final dimension-reduced hyperspectral image data; taking any one pixel point in the final dimension-reduced hyperspectral image data as a center point, and taking a neighborhood of different scales along all spectral dimensions to obtain a neighborhood pixel block of different scales; classifying the hyperspectral image according to the neighborhood pixel block of different scales by using a trained hyperspectral image classification model; and outputting a classification result; wherein the hyperspectral image classification model comprises: a plurality of 3DCNN layers, a convolution block attention weight module, and a fully connected layer; the present application considers the spectral features, spatial features and spatial spectral information of the hyperspectral image data, and calculates the weight of the spatial spectral information through the convolution block attention weight module, thereby improving the accuracy of the hyperspectral image classification.

This invention discloses a hyperspectral image unmixing method based on a three-dimensional deformation collaborative unmixing network. The method includes: constructing a hyperspectral image unmixing network comprising a three-dimensional joint feature encoding module, a physically constrained abundance mapping module, and a linear spectral reconstruction module; the encoding end rearranges the input hyperspectral image into a five-dimensional tensor with the spectral dimension as the depth dimension, and performs multi-scale spatial-spectral feature extraction through multiple three-dimensional deformation collaborative feature blocks; after deep feature extraction, the physically constrained abundance mapping module generates an abundance map satisfying the abundance non-negativity constraint and the abundance sum of one constraint; and the linear spectral reconstruction module reconstructs the hyperspectral image and outputs the endmember spectrum. This invention employs an end-to-end unsupervised training method, which can improve the endmember extraction accuracy and abundance estimation accuracy of hyperspectral images in complex scenes, low signal-to-noise ratio scenes, and irregular target regions.

The application belongs to the field of deep learning and hyperspectral imaging, and discloses a hyperspectral land cover classification method based on a three-dimensional dense connection network, which comprises the following steps: step 1) obtaining a data set; step 2) performing standardization processing on a spectral dimension; step 3) constructing a classification model based on hyperspectral imaging technology and deep learning; step 4) training and optimizing the model; and step 5) saving the model and applying it; the classification model integrates a parallel learnable attention module in a dense block in a DenseNet3D model; the parallel learnable attention module comprises an efficient channel attention mechanism module and a grouped channel attention mechanism module, and efficiently calculates channel attention and spectral grouped attention in parallel, and performs adaptive weighted fusion through a learnable weight fusion subnetwork, and finally outputs a feature map that is enhanced in the spectral and spatial dimensions. The application simultaneously improves the extraction and representation ability of discriminative features and the extraction ability of spectral-spatial joint features.

The application provides a plastic packaging identification method and system, electronic equipment and storage medium. The plastic packaging identification method comprises the following steps: obtaining hyperspectral data of plastic packaging, image data of plastic packaging and conveying speed of plastic packaging; constructing the hyperspectral data into a hyperspectral data cube according to the conveying speed, and performing spectral dimension standardization processing on the hyperspectral data cube to generate a target spectrum vector; correcting the image data to generate visual model data; performing spectral feature matching on the target spectrum vector to generate a material category label; performing geometric feature solving on the visual model data to generate a geometric attribute label; mapping the geometric center coordinates corresponding to the geometric attribute label to a coordinate system in which the hyperspectral data cube is located; and generating identification data packets under the condition that the material category label and the geometric attribute label are spatially consistent.

This invention provides a hyperspectral image enhancement method based on a stereo window Transformer, comprising: acquiring a first hyperspectral image to be processed; processing the first hyperspectral image using a trained stereo window Transformer network to obtain a second hyperspectral image, wherein the spatial resolution of the first hyperspectral image is lower than that of the second hyperspectral image; the stereo window Transformer network includes a shallow feature extraction unit, a deep feature extraction unit, and a high-resolution reconstruction unit; the deep feature extraction unit is used to jointly model the spatial and spectral three-dimensional dependencies of the image, and dynamically adjusts the window size in conjunction with a hierarchical heterogeneous strategy to adapt to different levels of spatial-spectral features. This invention utilizes a serial stereo interactive architecture to combine spatial and spectral dimensions, establish a three-dimensional stereo dependency, and realize the explicit transfer of spectral information to the spatial dimension; it can significantly improve the texture details and spectral fidelity of the reconstructed image, obtaining a higher quality super-resolution image.

The application discloses an oil and gas pipeline leakage intelligent detection and early warning positioning method, which comprises the following steps: acquiring pipeline vibration wave, pressure change and temperature field signals along the line through multi-dimensional signal cooperative acquisition, inputting the signals into a spectral space attention detection network for spatial and spectral dimension joint feature mining, and generating a high-dimensional leakage feature vector; calling a ground penetrating radar signal inversion model to analyze the change of medium electromagnetic parameters around the pipeline, positioning the preliminary range of the potential leakage area, and simultaneously establishing a pressure propagation and distance mapping relationship by using a pressure curve distance positioning algorithm; inputting the inversion result and the positioning result into a pipeline leakage intelligent early warning decision platform for multi-source data fusion verification, outputting accurate leakage position coordinates and leakage degree characteristic parameters, and forming a complete technical link from signal acquisition, feature mining, inversion positioning to fusion decision, so that the accuracy of leakage detection and the timeliness of early warning response are improved.

This invention discloses a multispectral imaging method and system based on precision micro-scanning and active spectral modulation. The system includes: a host computer module for sending commands to synchronously control spectral switching and micro-displacement of the image sensor, and for receiving acquired multispectral data; a spectral image acquisition module, consisting of an industrial telecentric lens, a liquid crystal tunable filter, an image sensor, and a precision micro-scanning stage, which acquires image sequences with sub-pixel displacement in specific spectral bands according to commands; and a spectral image reconstruction module that uses a super-resolution algorithm driven by both physics and data to process the image sequences in each band and generate a high-fidelity multispectral image cube. This invention, by integrating active spectral modulation and sub-pixel displacement technologies, solves the problem of simultaneously achieving high spectral dimension, high resolution, and high geometric fidelity in multispectral imaging, providing a solution for semiconductor detection, biomedicine, and other fields that combines ultra-high spatial detail with accurate spectral information.

The invention discloses a rapid wave band screening and dimension reduction method and system for a hyperspectral data cube, particularly relates to the technical field of data processing, and is used for solving the problem that in the prior art, processing efficiency and effective feature fidelity are difficult to consider in the hyperspectral data dimension reduction process. The method comprises the following steps: acquiring a hyperspectral data cube to be processed, performing spectral dimension correlation analysis on the hyperspectral data cube to generate an inter-band correlation matrix, and clustering all bands based on the matrix to obtain a plurality of band clusters; then, the response matching degree of wave bands in each wave band cluster is calculated according to the typical spectral response interval of the target geological object, the optimal representative wave band of each cluster is screened out, and then the cross-cluster characteristic complementarity of the optimal representative wave bands is evaluated to eliminate characteristic repeated wave bands; and finally, integrating the screened optimal representative wave band set to form a dimension-reduced hyperspectral data subset. The data processing efficiency can be ensured, and meanwhile, key spectral characteristics with indicating significance on a geological exploration target can be effectively reserved.