47 results about "Spectral vector" patented technology

The invention relates to the technical field of medicine quality detection, and particularly discloses an optical detection method and system for the content of vitamin tablets. The method comprises the following steps: driving an optical fiber probe array to carry out three-dimensional multi-angle near-infrared scanning through a multi-axis mechanical arm, and collecting reflection spectrum data; a weighted spectrum is generated through derivative spectrum analysis and double evaluation, and auxiliary material noise is corrected and separated in combination with subspace projection and a dynamic kernel function; performing spectral vector projection and regression coefficient iterative optimization through an online PLS model, and synthesizing an error coefficient based on a four-level index retrieval standard library to perform dual-channel feedback; and finally, fusing the credibility weight to output a detection result of the binding confidence. The method realizes nondestructive and high-precision detection of the vitamin tablets, has the beneficial effects of multi-dimensional data fusion, dynamic error compensation and high model adaptability, and remarkably improves the detection accuracy and reliability.

The invention provides a germplasm vigor detection method, device and equipment based on hyperspectral imaging and a medium, and relates to the technical field of intelligent agriculture, and the method comprises the following steps: obtaining spectral data of a target seed; determining a hyperspectral data cube of the target seed based on the spectral data; preprocessing the hyperspectral data cube to obtain a spectral vector corresponding to the hyperspectral data cube; extracting a local convolution feature of the spectral vector and a global feature of the hyperspectral data cube, and fusing the local convolution feature and the global feature to obtain a germplasm fusion feature; inputting the germplasm fusion features into a pre-trained classifier to obtain a target seed vigor detection result output by the classifier; the classifier is obtained based on germplasm sample features and corresponding label training. According to the method, the germplasm fusion features are obtained through the spectral data, so that the activity detection result of the target seeds is obtained, destructive experiments on the seeds are not needed, the detection efficiency and portability are improved, and the labor cost is reduced.

The invention relates to the technical field of foreign matter recognition, in particular to a foreign matter early warning method and device, electronic equipment and a storage medium. Carrying out dimensionality reduction on the plurality of first spectral vectors through a dimensionality reduction conversion matrix to obtain a plurality of second spectral vectors; classifying each second spectrum vector according to the relationship between the second spectrum vectors and a plurality of spectrum classes to obtain a plurality of spectrum class identifiers; and finally, selecting a plurality of target identifiers from the plurality of spectrum identifiers, determining a first foreign matter probability according to the plurality of target identifiers, the perimeter area ratio of the first region and a conditional probability equation, and performing early warning according to the first foreign matter probability. According to the invention, based on the visual identification area, the spectral data is extracted, dimensionality reduction and classification are carried out on the spectral data, the foreign matter risk is determined based on data dimensionality reduction, classification and statistics, the accuracy of foreign matter early warning is improved, and the probability of false alarm when the foreign matter is found is reduced.

The invention discloses a hyperspectral unsupervised anomaly detection method based on course learning and spectral parity decomposition. The method comprises the following steps of: 1, preprocessing an input hyperspectral image to obtain a spectral vector, and constructing a region-level training sample; 2, constructing a hyperspectral anomaly detection model based on curriculum learning and spectral parity decomposition, and obtaining representative spectral features of the obtained fine-grained region; 3, defining a mean square error loss function of the hyperspectral anomaly detection model; 4, training the hyperspectral anomaly detection model; 5, reconstructing the representative spectral features of the fine-grained region by using the trained hyperspectral anomaly detection model to obtain a reconstructed image, and calculating a reconstruction error graph of the spectral vector and the reconstructed image; and 6, carrying out binarization segmentation on the reconstructed error image to obtain an anomaly detection result. According to the method, the false alarm rate and the calculation complexity are reduced while the high detection precision is ensured, and the dual requirements of an actual scene for efficiency and reliability are met.

The invention discloses a plant leaf physicochemical parameter inversion method based on spectrum two-dimensional characterization and multi-task deep learning, and belongs to the technical field of plant physiological detection. The method comprises the following steps: constructing a hyperspectral plant leaf data set; preprocessing a hyperspectral plant leaf data set; reconstructing the preprocessed one-dimensional spectral vector into a two-dimensional image; taking the generated two-dimensional image as input, constructing a multi-task learning model, and performing training and evaluation; and applying the trained model to a test set to obtain an inversion result of the physical and chemical parameters of the leaf. The method solves the problem of parameter inversion interference caused by the same-spectrum foreign matter phenomenon, achieves the precise modeling of local fine feature extraction and global context relation, solves the inversion precision bottleneck caused by spectral response overlapping among parameters, improves the distinguishing capability of the model for easily confused parameters such as chlorophyll and carotenoid, and improves the accuracy of the model. And finally, high-precision and robust synchronous inversion of the physical and chemical parameters of the plant leaves is realized.

The invention discloses a perception method based on multi-semantic space attention and Daubechies wavelet double-branch Mama, and belongs to the field of hyperspectral image processing. The problem that in the existing Mama-based model feature extraction process, the capacity of capturing multi-scale local structures and direction sensing information is insufficient is solved. The method comprises the following steps: inputting a hyperspectral image; projecting the spectral vector to an embedding space through an embedding layer to obtain an embedding feature; inputting the embedded features into an encoder, wherein the encoder comprises an SMSAMama branch, a DWTMama branch and a self-adaptive feature fusion module; the SMSAMama branch is used for extracting spatial features; the DWTAMba branch is used for extracting spectral features; the adaptive feature fusion module performs weighted integration on the spatial features and the spectral features by using randomly initialized fusion weights; and inputting the integrated features into a segmentation head to generate a final perception result. The method is used in agricultural monitoring and urban planning fields.

The invention provides a hyperspectral spectrum data fusion method and a hyperspectral spectrum data fusion system capable of widening a spectral band, and the method comprises the steps: carrying out the wavelength domain matching of Vis-NIR hyperspectral data obtained by a ground object spectrometer based on the visible light band of a hyperspectral image, and storing the resampled Vis-NIR hyperspectral data in a spectrum library. Performing similarity measurement on the spectral vector of each pixel in the hyperspectral image and all standard spectrums in the spectrum library to generate a spectral feature fingerprint map and a spectral response coefficient map; and performing spectrum reconstruction on each pixel in the hyperspectral image data in combination with the spectral feature fingerprint map and the spectral response coefficient map. And finally, obtaining a hyperspectral image cube after broadened spectrum fusion. According to the scheme, on the basis of wavelength domain matching, the spectral feature fingerprint map, the spectral response coefficient map and the like, the hyperspectral image cube obtained through final fusion covers a wider spectral range, and the original high spatial resolution is reserved.

The invention relates to the technical field of foreign matter recognition based on hyperspectrum, in particular to a foreign matter recognition method and device based on spectral feature selection, equipment and a medium, and the method comprises the steps: firstly obtaining a main dimension conversion matrix and a plurality of first dimension conversion matrixes; converting each first spectral vector in the first spectral vector set according to the main dimension conversion matrix, and screening out a plurality of first risk spectral vectors from the first spectral vector set according to a conversion result; performing conversion on each first risk spectral vector through the plurality of first dimension conversion matrixes, and performing spectral characteristic matching according to a conversion result; and finally, extracting a first image contour from the target image according to spectral characteristics, and sending the first image contour to an identification model for foreign matter identification. The method is good in preprocessing effect, and the recognition efficiency and the recognition precision are improved.

The application discloses a hyperspectral image real-time local anomaly detection method based on multi-line wave band processing, which comprises the following steps: reading hyperspectral image data, initializing the number of wave bands n determined by the number of filters, initializing the number of lines j based on wave band extraction, and initializing the interval number k between different wave bands; converting the data into data required for real-time detection processing according to the values of the number of wave bands n, the number of lines j and the interval number k; establishing a state equation of a spectral vector correlation matrix R(n) of hyperspectral data pixels; updating the inverse matrix R(n) of the correlation matrix R(n) by using a block matrix inversion formula ‑1 ; detecting the hyperspectral data required for real-time detection processing by using a multi-line wave band correlation matrix real-time local anomaly RX operator, and finally obtaining the result of the hyperspectral image real-time local anomaly detection. The application realizes local anomaly detection of the hyperspectral image, avoids high-dimensional data storage and repeated calculation, and has a good real-time local anomaly detection effect.

This invention provides a 3D printing metal material grinding device and grinding system, comprising: acquiring a sequence of hyperspectral images of metal powder particles to be ground, identifying and segmenting individual particles; extracting the pixel coordinates of the particle edge contour and the corresponding spectral response vector, and normalizing the spatial coordinates and spectral vectors respectively; constructing a first covariance matrix representing the spatial distribution of particles and a second covariance matrix representing the material distribution; performing singular value decomposition on the two matrices respectively, terminating the iteration when the relative change of the principal singular values ​​is less than a preset threshold, and obtaining singular values ​​and singular vectors; using the ratio of the two largest principal singular values ​​of the first matrix as the contour anisotropy parameter; using the sum of the singular values ​​of the second matrix plus the reciprocal of a preset constant as the material uniformity parameter; inputting the two types of metric parameters into a pre-trained control model, generating control commands and sending them to the grinding equipment actuator.

The present application relates to the technical field of foreign matter identification, and particularly relates to a foreign matter early warning method and device, electronic equipment and storage medium, the method of the present application first acquires a first region; then a plurality of first spectral vectors are reduced in dimension through a dimension reduction conversion matrix to obtain a plurality of second spectral vectors; then each second spectral vector is classified according to the relationship between the second spectral vector and a plurality of spectral classes to obtain a plurality of spectral class identifiers; finally, a plurality of target identifiers are selected from the plurality of spectral class identifiers, a first foreign matter probability is determined according to the plurality of target identifiers, a perimeter area ratio of the first region and a conditional probability equation, and early warning is performed according to the first foreign matter probability. Based on the region of visual identification, spectral data is extracted, the spectral data is reduced in dimension and classified, the foreign matter risk is determined based on data dimension reduction, classification and statistics, the accuracy of foreign matter early warning is improved, and the probability of false alarm when discovering foreign matter is reduced.

The invention provides a hyperspectral image aerial small target detection method and device, and the method comprises the steps: determining a plurality of wavebands with the maximum contrast based on a local contrast feature pattern of each waveband in a target hyperspectral image, and obtaining the target hyperspectral image after the waveband selection; calculating the enhanced mahalanobis distance of the spatial-spectral variance ratio of each spectral vector in the target hyperspectral image after the band selection; and determining a small target detection result of the target hyperspectral image based on the Mahalanobis distance with the enhanced spatial-spectral variance ratio. Therefore, small targets in the air can be monitored, wavebands with differences between the targets and the background can be well screened out, redundant background interference is reduced, the detection precision is improved, and the false alarm rate is reduced.

This invention provides a method and system for quality detection of edible fungi based on spectral features, belonging to the field of spectral detection technology. This invention acquires the light intensity sequence of a detection light source under a preset fluctuation state, extracts standard deviation, root mean square, and frequency features to construct a multidimensional fluctuation feature vector, achieving a comprehensive characterization of the light source's fluctuation characteristics. This provides rich input information for the offset prediction model, significantly improving the accuracy of spectral correction. Using the multidimensional fluctuation feature vector as the independent variable and the spectral offset as the dependent variable, an offset prediction model corresponding to each wavelength is established. Target wavelengths are then selected based on the goodness of fit, achieving high-precision correction of the original spectral data. The corrected spectral data is compared with the grade template spectral data in the quality template library using cosine similarity calculation, and the quality grade is selected by spectral vector matching, thereby significantly improving the stability and reliability of the edible fungi quality detection results.

The invention discloses a multi-dimensional light field information space construction method for high-resolution infrared imaging, and the method comprises the steps: selecting an imaging scene, and determining a spectrum scanning voltage range; fixing a polarization angle, carrying out spectral voltage imaging scanning, carrying out vector slicing on an image obtained by each obtained spectral scanning voltage, stacking into a column vector, and finally obtaining a two-dimensional spectral vector space; the polarization angles are rotated, spectral voltage imaging scanning continues to be carried out, and a two-dimensional spectral vector space is obtained when each polarization angle is rotated; after all collection is finished, combining the two-dimensional spectrum vector spaces obtained at each polarization angle together in sequence to form a multi-dimensional light field information space containing a space image, spectrum information and polarization information; according to the invention, spectrum scanning is controlled through voltage capable of in-situ regulation and control, a complete multi-dimensional light field information space containing a space image, a spectrum and polarization is constructed, and high-resolution infrared imaging is realized.

The invention relates to the technical field of culture medium detection, and discloses a fruit fungus culture medium pollution identification method and system based on hyperspectral imaging, and the method comprises the steps: obtaining a data cube of a hyperspectral image of a culture medium, executing non-negative tensor decomposition, and obtaining a group of space factor graphs and corresponding spectral vectors; calculating a Bhattacharyya distance between each spectrum vector and a preset pollution strain spectrum library, and generating a pollution affinity coefficient corresponding to each space factor graph; calculating the mahalanobis distance of each pixel point in the data cube, and generating a global spectral distortion graph; initializing a pollution potential field with the same spatial size as the hyperspectral image, and carrying out iterative updating through a reaction diffusion equation until convergence; and performing threshold segmentation on the converged pollution potential field to obtain a binary mask of the pollution area. According to the method, the analysis of the spectral dimension and the structural constraint of the spatial dimension are deeply fused by constructing the reaction diffusion model, so that the early and accurate identification of the culture medium pollution is realized.

The application provides an online water quality analysis method and system in a radioactive waste liquid treatment process, and relates to the technical field of online water quality monitoring.The method comprises the following steps: step 1, passing a clarified sample into an analysis pool, collecting Raman spectral characteristic peak data in real time through a plurality of groups of optical fiber Raman probe arrays arranged in the analysis pool along the flow direction of the waste liquid; performing principal component analysis on the Raman spectral characteristic peak data, extracting characteristic spectral vectors corresponding to main radioactive nuclides and interfering substances, and constructing a two-dimensional concentration distribution field according to the continuous change of the characteristic spectral vectors in the spatial position; and step 2, extracting an isochromatic line and performing discrete sampling on a specific interfering substance component in the two-dimensional concentration distribution field, calculating the minimum circumscribed circle of all sampling points, and obtaining the center coordinates and radius of the isochromatic line.The application realizes intelligent regulation and control of treatment process parameters and guarantees that the discharged water quality is stable and up to the standard.

The present application belongs to the technical field of remote sensing, and relates to a weight dictionary thermal infrared hyperspectral mixed pixel temperature emissivity solving method. The method comprises: obtaining simulation data and performing inversion channel screening according to the simulation data; obtaining metadata and quality control marks and performing effective pixel screening; performing atmospheric correction; constructing an initial value dictionary and training to obtain an emissivity dictionary; calculating a ground surface emissivity spectral vector, reconstructing the emissivity spectral vector based on a ROMP algorithm, performing weight redistribution through a weight distribution algorithm, and iteratively solving to restore the ground surface temperature and the noise-free and unbiased emissivity spectrum of the mixed pixel. The present application combines the ROMP algorithm and the weight distribution algorithm into a WOMP algorithm, overcomes the greediness of the traditional algorithm, cooperates with the initial value dictionary and the emissivity dictionary, reconstructs the unbiased emissivity spectrum, realizes the synchronous inversion of the ground surface temperature and the emissivity spectrum, and significantly improves the stability, unbiasedness and decoupling capability of the mixed pixel inversion.

The invention provides a fire range extraction method based on a nuclear combustion index and adaptive spatial spectrum optimization, and relates to the technical field of remote sensing image processing, and the method comprises the steps: collecting remote sensing image data; mapping a spectral vector corresponding to the fire image data set to a high-dimensional space based on a Gaussian radial basis kernel function, and performing kernel principal component analysis on the spectral vector to construct a kernel difference combustion index; according to kernel density estimation, performing adaptive threshold segmentation on the kernel difference combustion index to obtain an initial fire pixel detection result; constructing a spectrum space composite kernel similarity matrix, and optimizing the initial fire pixel detection result in a conditional random field according to the spectrum space composite kernel similarity matrix to obtain a transition fire pixel detection result; and performing morphological operation and small patch area filtering on the transition fire pixel detection result to obtain a predicted fire pixel detection result. According to the invention, the overall precision of fire range extraction can be improved.

The invention relates to the technical field of remote sensing image processing and ecological environment monitoring, and discloses a remote sensing image batch index calculation and trend analysis system and method based on MATLABGUI. The system comprises a data acquisition and metadata analysis module, a heterogeneous data adaptive mapping module, a multi-dimensional feature matrix construction module, a time sequence trend mining module and a visual interaction module. Remapping the heterogeneous source data into a standard logic spectrum vector by utilizing a wave band mapping lookup table; calculating ecological characteristic components in parallel by adopting a dynamic calculation graph and a matrix operator, and analyzing and synthesizing a remote sensing ecological index through a principal component corrected by characteristic polarity; and carrying out vectorization Sen slope estimation and a Mann-Kendall significance test. According to the method, through a main thread and background task thread separation architecture, multi-source data compatibility, calculation process self-adaption fault tolerance and efficient mining of a long time sequence trend are achieved.

The invention discloses a hyperspectral image reconstruction method and system, and the method comprises the steps: obtaining a hyperspectral image, obtaining a spectral matrix of the hyperspectral image, and enabling any line of vectors of the spectral matrix to be spectral vectors of corresponding pixels in the hyperspectral image; inputting the spectrum matrix into a preset encoder, and outputting a low-dimensional representation of each row of vectors of the spectrum matrix; a first preset image is constructed, data of any pixel of the first preset image is formed by low-dimensional representation of a corresponding row vector in the spectrum matrix, the first preset image is input into a preset model, a second preset image is output, and the resolution ratio of the second preset image is higher than that of the first preset image; the second preset image is input into a preset decoder, a third preset image is output, the number of channels of data of any pixel of the third preset image is larger than the number of channels of low-dimensional representation of the corresponding row vector in the spectrum matrix, and a reconstructed hyperspectral image is obtained according to the third preset image. According to the invention, the hyperspectral image with improved spatial resolution can be obtained.

This invention discloses a noise detection method. The method and readable storage medium include: digitizing the audio to be detected and normalizing it to form a time-domain audio vector; performing a Fast Fourier Transform on the time-domain audio vector data, adding each spectral vector to a queue, calculating the similarity index and power dispersion of the first spectral vector in the queue compared to other spectral vectors, and calculating a noise index; determining whether the noise index of the first spectral vector belongs to a noise range, and if so, identifying the audio corresponding to the first spectral vector as noise data. In this invention, based on the highly dispersed spectral distribution and temporal stability of noise, the similarity index and power dispersion are combined into a noise index that can distinguish noise from normal audio, thereby accurately identifying noise signals in broadcast signals.

The invention discloses a hyperspectral anomaly detection method based on adaptive superpixels and spatial spectrum constraints. The method comprises the following steps: firstly, inputting a hyperspectral image Y into entropy rate optimization superpixel segmentation to obtain a kth superpixel image Pk; calculating a super-pixel block change measure psi (Pk) of the Pk, calculating a change weight criterion of the psi (Pk), and comparing the change weight criterion with a scale selection threshold rho to obtain an optimal super-pixel image Popt; constructing a super-pixel block subset S [alpha] by using the Popt and the quantile [alpha], determining a super-pixel mask M of the S [alpha], and constructing a super-pixel adjacent region fNbr (q) through the ith super-pixel block qi in the Popt and the S [alpha]; calculating an average spectral vector mu * through the pixel spectral vector of the fNbr (qui), and obtaining an image Y * after background filling by using M and mu *; setting an outer window area A and an inner window area B in the Y * so as to construct a background dictionary D; calculating a superpixel difference score gamma by using the pixel number nopt (q) of qi, constructing a weight matrix omega by using gamma, and determining an optimal representation coefficient beta; and finally, calculating by using D and beta to obtain a hyperspectral anomaly detection result graph R.

The invention discloses a fermentation state recognition method and system for straw feed, and relates to the technical field of state recognition, full-band hyperspectral image data and RGB image data of a straw feed sample are collected, key characteristic wave band data are obtained based on the full-band hyperspectral image data, and the key characteristic wave band data and the RGB image data form a multi-modal original database; screening to obtain a pH sensitive characteristic wave band subset, constructing a spectral acidity space mapping model, carrying out pH value inversion on pixel points to generate a pH value space distribution thermodynamic diagram, obtaining an abnormal region mask, mapping the abnormal region mask back to a multi-modal original database, obtaining a to-be-detected target data block, and obtaining the local moisture content of an abnormal region based on key wave band reflectivity data. Extracting a robust visual feature vector and an organic acid feature spectrum vector from the to-be-detected target data block, constructing a moisture correction feature vector in combination with the local moisture content to obtain the acetic acid content and the propionic acid content, and identifying the fermentation state in combination with the abnormal region mask and the local moisture content.

This application relates to the field of intelligent detection technology, and in particular to a method and system for detecting mycotoxins in feed based on spectral analysis. The method includes: acquiring the near-infrared diffuse reflectance spectrum of the feed sample to be tested; constructing an original spectral vector based on the absorbance data of the near-infrared diffuse reflectance spectrum; determining the projection component of the original spectral vector on a pre-constructed qualified matrix space; subtracting the projection component from the original spectral vector to obtain a residual vector; calculating the first-order difference spectrum of the residual vector based on preset sensitivity weights; determining weighted eigenvalues ​​by local feature weighting of the first-order difference spectrum; determining the residual entropy value and the range value of the residual vector; and using the weighted eigenvalues, residual entropy value, and range value to determine the detection result of the feed sample to be tested. Through the above technical solution, more accurate detection results for mycotoxins in feed can be obtained.

This invention discloses an audio information processing method and system, applied in the field of audio signal processing technology. The method includes extracting spectral features from audio data to be processed and performing spectral encoding on the spectral features to obtain a spectral vector representation; performing temporal encoding on the audio data to be processed to obtain a temporal vector representation; obtaining a first voiceprint result based on the spectral vector representation and the temporal vector representation; and obtaining a second voiceprint result based on the spectral vector representation and the temporal vector representation; integrating the first and second voiceprint results into a voiceprint embedding set; performing max pooling on the voiceprint embedding set to obtain a voiceprint feature embedding vector; and determining the voiceprint recognition result of the audio data to be processed based on the analysis results of the voiceprint feature embedding vector. The method of this invention improves the accuracy of voiceprint recognition results.

This invention discloses a method for predicting and correcting the uniformity of PE film thickness, belonging to the field of prediction and correction technology. It addresses the problem of unmet requirements for high stability and high consistency in film thickness control. By setting a monitoring time window, lateral thickness data of the PE film is collected within the time window. The thickness periodic characteristics are evaluated and thickness disturbance states are classified. Based on the classification results, a fluctuation analysis mechanism is triggered to acquire screw speed data and perform Fourier transform. Coupled with the thickness data, the coupling spectral density is calculated. Melt pressure and traction tension amplitude data are collected to construct spectral vector features. The dominant frequency peak is extracted to predict the time-domain disturbance, which is then corrected based on the coupling spectral density. The thickness deviation is judged based on the corrected time-domain disturbance, and the die head adjustment displacement is generated. During the correction process, the thickness change rate is collected to determine the correction status, improving the accuracy of thickness deviation identification and adjustment response precision, and enhancing the stability and consistency of film thickness control.

This invention discloses a DOA super-resolution estimation method and product based on Toeplitz covariance enhancement and neural networks. The method includes: acquiring received data from a uniform linear array, calculating the sample covariance matrix, and performing trace normalization; performing Toeplitz prior correction by averaging the elements of each diagonal and backfilling to construct the Toeplitz prior covariance, and performing Hermitization; based on the sample covariance matrix and the Toeplitz prior covariance, adopting a linear shrinkage model and determining the optimal shrinkage coefficient using the Frobenius norm mean square error as the criterion to obtain the shrinkage covariance matrix; performing vectorized projection onto an overcomplete array dictionary to obtain complex projection features; splitting the features into real and imaginary parts to form a dual channel, inputting it into a convolutional neural network to obtain a spatial spectral vector; and performing spectral peak search to extract the angles corresponding to the K peaks with the largest amplitudes to obtain the DOA estimation result.

This invention discloses a real-time anomaly detection method for hyperspectral data based on a finite-order Markov model, comprising: reading hyperspectral image data and initializing the number of rows; establishing a finite-order Markov model of inter-row information, calculating the information content of the current row i, and calculating the second derivative of the information content to determine whether the current row is an anomaly or a non-anomaly; establishing the state equation of the spectral vector correlation matrix R(n) of the hyperspectral data pixels; if it is an anomaly, replacing the first row of the original data with new data in the spectral vector correlation matrix R(n); if it is a non-anomaly, adding new data to the original data; and updating the inverse matrix R of the correlation matrix R(n). ‑1 (n); This invention performs detection on hyperspectral data and obtains the results of real-time local anomaly detection in hyperspectral images. It achieves local anomaly detection in hyperspectral images, avoids high-dimensional data storage and redundant computation, and has good local anomaly detection performance.

The present application relates to the technical field of foreign matter identification based on hyperspectrum, and particularly relates to a foreign matter identification method and device based on spectral feature selection, equipment and medium. The method of the present application first acquires a main dimension conversion matrix and a plurality of first dimension conversion matrices. Then, for each first spectral vector in a first spectral vector set, the main dimension conversion matrix is used for conversion, and a plurality of first risk spectral vectors are selected from the first spectral vector set according to the conversion result. Next, for each first risk spectral vector, the plurality of first dimension conversion matrices are used for conversion, and spectral characteristics are matched according to the conversion result. Finally, a first image contour is extracted from the target image according to the spectral characteristics, and the first image contour is input into an identification model for foreign matter identification. The present application has good pretreatment effect, and improves the identification efficiency and accuracy.