A Hyperspectral Anomaly Detection Method Based on a Dual-Line Iterative Convolutional Neural Network

Abstract

The present invention discloses a hyperspectral anomaly detection method based on a dual-line iterative convolutional neural network, which includes the following steps: Step S1, input and parameter setting; Step S2, anomaly enhancement line; Step S3, background reconstruction line; Step S4, anomaly point determination; Step S5, hyperspectral data update; Step S6, detection map generation. In the present invention, the background reconstruction line focuses on background reconstruction, and the anomaly enhancement line is responsible for anomaly enhancement. The two complement each other and promote each other, realizing more accurate anomaly detection and overcoming the problem that it is difficult to effectively combine background reconstruction and anomaly detection in traditional methods. In the present invention, combined with the superpixel technology, an anomaly detection strategy based on comparison with surrounding pixel points is used, which improves the sensitivity to local areas, makes the selection of anomaly targets simpler and more efficient, improves the detection efficiency, and avoids possible computational redundancy and efficiency bottlenecks.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral anomaly detection, and more specifically, to a hyperspectral anomaly detection method based on a dual-line iterative convolutional neural network. Background Art

[0002] In recent decades, researchers have proposed many effective hyperspectral anomaly detection algorithms. These hyperspectral anomaly detection (HAD) algorithms can be classified into algorithms based on statistical metrics, representation models, and deep learning.

[0003] A. Anomaly Detection Based on Statistical Metrics

[0004] One of the most well-known anomaly detection methods is the Reed-Xiaoli (RX) detector, which was proposed by Reed and Yu. It is a distance-based anomaly detection method widely used in the anomaly detection task of hyperspectral images (HSIs). The RX algorithm is based on the generalized likelihood ratio test. It models the background as a multivariate Gaussian distribution and uses the mean vector and covariance matrix of background pixels to describe the statistical characteristics of the background. It judges the anomaly degree of the pixel to be detected by calculating the Mahalanobis distance between the pixel to be detected and the background mean. The RX detector is divided into global RX and local RX. The global RX detector assumes that the background in the data has statistical uniformity and uses the entire data set to estimate the mean and covariance matrix of the background, thereby constructing a global background model. The local RX is based on local background information, selects a neighborhood window around the target pixel, and estimates the mean and covariance matrix of this area as the background model, which can detect local anomalies more sensitively. Based on this idea of the RX algorithm, some improved methods have been proposed in subsequent research, such as kernel RX (KRX), which processes nonlinear data by introducing a kernel function, projects the original image into a nonlinear high-dimensional feature space, and considers the higher-order characteristics of different feature bands. The anomaly detection method based on the RX algorithm is limited by the Gaussian distribution assumption and the high computational complexity of inverting the high-dimensional covariance matrix. To solve the above problems, Schweizer et al. proposed an adaptive detection algorithm based on a three-dimensional Gaussian Markov random field, which uses a spatio-spectral joint random field to characterize the background statistical characteristics and combines spatial and spectral information to alleviate the computational cost of covariance inversion; Banerjee proposed a method based on support vector data description (SVDD) to detect target samples through nonlinear projection.

[0005] B. Anomaly Detection Based on Representation Models

[0006] The anomaly target detection method based on the representation model does not rely on the probability distribution assumption of the background. Instead, it designs the algorithm optimization criterion by analyzing the local spectral distribution characteristics or applying various knowledge constraints, uses a set of spectral vectors to describe the background and anomaly targets, and judges the anomaly degree of the samples through the residual value. Li et al. proposed the collaborative representation-based detector (CRD), which considers the information of neighboring pixels to represent the central pixel, improving the detection accuracy and robustness. In addition, in the abundance and dictionary-based low-rank decomposition (ADLR) method, it is assumed that the background after unmixing shows low-rank characteristics in the abundance subspace. Xu et al. proposed the low-rank and sparse representation (LRASR) method, which uses the low-rank and sparse representation method to model the background as a low-rank matrix and the anomaly target as sparse noise, enhancing the ability to separate anomaly points. At the same time, since the background can be approximately represented by a set of basis background signals, the low-rank prior is applied for background estimation in hyperspectral anomaly detection. The low-rank sparse matrix decomposition (LRaSMD) uses the Mahalanobis distance method (LSMAD) to decompose the data in the hyperspectral image into a low-rank background, sparse anomaly, and noise parts. However, most methods only consider the spectral characteristics of the background and anomaly components, while ignoring their spatial distribution characteristics.

[0007] C. Deep Learning-Based Anomaly Detection

[0008] With the rapid development of deep learning technology, hyperspectral anomaly detection methods based on deep learning have gradually become a research hotspot. Some improved methods have begun to fuse multiple deep learning models to improve the detection accuracy and robustness. For example, based on the property that graph regularization can preserve local spatial features in images, Fan et al. proposed the Robust Graph Autoencoder (RGAE), adding graph regularization to the AE to improve the network's robustness to abnormal targets and noise. Wang et al. proposed a method based on the fully convolutional autoencoder, performing unsupervised learning on hyperspectral images through the autoencoder to obtain an anomaly probability map, and then using an adaptive weighted loss function to guide anomaly detection. This method can suppress abnormal reconstruction. Therefore, in the reconstruction error map, anomalies have a high contrast with the background. The weights are derived from the reconstruction error and can be adaptively updated. In addition, Generative Adversarial Networks (GANs) have also been widely applied to hyperspectral anomaly detection. Jiang et al. proposed an anomaly detection method based on GAN, using GAN to evaluate background statistical information and obtaining spectral features in a new semi-supervised spectral learning, which can more accurately identify anomaly points. GANs can learn complex distribution features from a large amount of normal data without relying on labels and help detect abnormal pixels through the generative network. Zhong et al. adopted a rough search strategy based on morphological filtering to extract a potential background subset and used it as the input for GAN training. Since the training samples consist only of background elements, abnormal samples cannot be accurately reconstructed in the test set, thus improving the model's ability to distinguish between background and abnormal samples. Wu et al. used a pre-trained convolutional neural network to detect the input image data, which can extract deep information features with strong robustness and high discriminability, and achieve the separation of abnormal targets from the image background. Zhang et al. proposed an adaptive CNN network, using tensor convolution and adopting a neighborhood tensor block as the convolution kernel to evaluate the statistical characteristics of test pixels, thereby mining effective features. To reduce the impact of non-stationary noise on the data, Zhang et al. proposed a tensor anomaly detection method combining FrFT and CNN architectures, enhancing the distinguishability between the background and anomalies in the frequency domain through tensor transformation and PCA preprocessing strategies.

[0009] Currently, the hyperspectral anomaly detection algorithms have the following defects:

[0010] (1) Noise sensitivity: Hyperspectral data may be affected by various noise sources during the acquisition process, such as sensor noise, environmental interference, etc.; the presence of noise may lead to false detections or missed detections, reducing the detection accuracy of the model;

[0011] (2) Limitations of background assumptions: Many hyperspectral anomaly detection methods usually rely on certain background assumptions. However, actual hyperspectral data is often complex and variable, and the background assumptions may not hold.

[0012] (3) High computational complexity: The processing of hyperspectral data often involves a large number of bands and pixels. If deep learning is used, the training time may be greatly increased.

[0013] (4) Poor generalization ability: Some hyperspectral anomaly detection methods perform well on some datasets, but their generalization ability is poor and they perform poorly on other datasets. Summary of the Invention

[0014] The purpose of the present invention is to overcome the above-mentioned deficiencies of the prior art and provide a hyperspectral anomaly detection method based on a dual-line iterative convolutional neural network.

[0015] To achieve the above purpose, the present invention adopts the following technical solutions:

[0016] A hyperspectral anomaly detection method based on a dual-line iterative convolutional neural network includes the following steps:

[0017] Step S1, input and parameter setting: Input hyperspectral data, input the hyperspectral data into the anomaly enhancement line and the background reconstruction line, and initialize the number of superpixels, the size of the convolutional kernel, the learning rate, and the number of iterations.

[0018] Step S2, in the anomaly enhancement line, input the hyperspectral data into the anomaly detector to obtain an anomaly probability map, and use connected component analysis to screen the anomaly regions to generate an anomaly detection map. Adopt a moving average update strategy to obtain an anomaly probability detection map.

[0019] Step S3, in the background reconstruction line, use a global robust CNN to perform superpixel segmentation on the hyperspectral data and construct a Laplacian matrix; fuse the above operations into the convolutional neural network to reconstruct the hyperspectral data, and use the error between the reconstructed data and the original data as the anomaly probability detection map.

[0020] Step S4, anomaly point determination: Multiply the anomaly probability detection maps obtained by the anomaly enhancement line and the background reconstruction line respectively to obtain the final anomaly probability map.

[0021] Step S5, hyperspectral data update: Update the hyperspectral data input to the anomaly enhancement line and the background reconstruction line in real time according to the final anomaly probability map.

[0022] Step S6, detection map generation: After the iteration is completed, use the difference between the input data of the updated background reconstruction line and the original data as the anomaly detection map for output.

[0023] Further, in step S3, the convolutional neural network adopts an encoder-decoder network structure. The encoder network is used to extract local features. After adding a Laplacian matrix to its output, it is input into the decoder network as the input of the decoder network.

[0024] Further, the steps for the encoder network to extract local features are as follows:

[0025] Step S31, represent the original hyperspectral data received by the network as , where m is the width of the image, n is the height of the image, and b is the number of bands of the image;

[0026] Step S32, in the encoder network, input the hyperspectral data into the input layer. In the first convolutional layer, perform convolution operations using 64 convolutional kernels of size 3×3×b, and output data after padding;

[0027] Step S33, transfer the output data to the second convolutional layer, perform convolution using 32 convolutional kernels of size 3×3×64, and output data after batch normalization and ReLU activation function.

[0028] Further, divide the hyperspectral data into M×M blocks, and represent the data of each block as , and apply the convolutional neural network independently on each block.

[0029] Further, in step S3, by introducing a superpixel segmentation algorithm, divide the hyperspectral data into fewer pixel groups with consistent regions, and use the modified Laplacian matrix to accelerate the construction of the Laplacian matrix. The objective function of the proposed convolutional neural network is as follows:

[0030] (1)

[0031] In formula (1), is the reconstruction error, is the Laplacian matrix regularization term, λ is the balance coefficient, N is the number of samples, and Z is the latent representation matrix of the hidden layer.

[0032] Further, step S2 specifically includes the following steps:

[0033] Step S21, domain difference analysis. Input the hyperspectral data into the anomaly detector. The hyperspectral data is fused by band and superpixel segmented. Each pixel point judges the anomaly degree according to the difference degree from the surrounding pixels to obtain the anomaly probability map, and screen the anomaly regions from the consistent image through connected region analysis, and remove the regions with too large area to generate the anomaly detection map;

[0034] Step S22, Moving Average Update: All anomaly detection maps are updated using a moving average update strategy to obtain an anomaly probability detection map.

[0035] Further, in step S2, the detection of the anomaly detector includes the following steps:

[0036] Step S211, Sum the hyperspectral data band by band to generate a grayscale image G;

[0037] Step S212, Represent each pixel value of the grayscale image G as the sum of all bands:

[0038]

[0039] Normalize the grayscale image G to the interval [0, 1];

[0040] Step S213, Divide the normalized grayscale image G into multiple superpixel regions through a superpixel segmentation algorithm, and assign a unique label to each superpixel , indicating the superpixel region to which the current pixel belongs;

[0041] Step S214, Let the label of the current superpixel be , and set a distance parameter d. Find the pixels surrounding the superpixel through distance neighborhood expansion. Let the current pixel value be , and the neighborhood pixel value be , and the formula for the neighborhood is:

[0042] = ;

[0043] Step S215, Calculate the mean difference between the current pixel value and the neighborhood pixel value :

[0044] ;

[0045] Step S216, Square the difference value and record it in the anomaly probability map M, and perform normalization processing:

[0046] ;

[0047] Step S217, Calculate the threshold T of the top 5% of the pixel values in :

[0048]

[0049] In the formula, Arrange all the pixel values in the anomaly probability map M in descending order, and convert the top 5% of the pixel values to 1;

[0050] Step S218: Perform connected region segmentation, calculate the number of pixel points contained in each abnormal component, and denote each pixel block as , i = 1, 2, 3...; If the number of pixel points of

[0051] is greater than 50, then it is considered that this pixel block is not an abnormal point and is denoted as 0; otherwise, retain the original value.

[0052] 1. In the present invention, the background reconstruction circuit focuses on background reconstruction, and the anomaly enhancement circuit is responsible for anomaly enhancement. The two complement each other and promote each other, achieving more accurate anomaly detection and overcoming the problem that it is difficult to effectively combine background reconstruction and anomaly detection in traditional methods.

[0053] 2. In the present invention, by combining the superpixel technology and using an anomaly detection strategy based on comparison with surrounding pixel points, the sensitivity to local areas is improved, the selection of abnormal targets is made more simple and efficient, the detection efficiency is enhanced, and possible computational redundancy and efficiency bottlenecks are avoided.

[0054] 3. In the present invention, a convolutional neural network is introduced into hyperspectral anomaly detection, and the global structural information of the Laplacian matrix is combined to achieve the extraction of multi-scale features, which not only improves the learning ability of the model for abnormal targets at different scales but also effectively enhances the adaptability and accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is an overall framework schematic diagram of the hyperspectral anomaly detection method based on a dual-circuit iterative convolutional neural network in this embodiment;

[0056] Figure 2 is a flowchart of the hyperspectral anomaly detection method based on a dual-circuit iterative convolutional neural network in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0058] Embodiment: A hyperspectral anomaly detection method based on a dual-circuit iterative convolutional neural network, as Figure 1 、 Figure 2 shown, includes the following steps:

[0059] Step S1, Input and parameter setting: Input hyperspectral data, input the hyperspectral data into the anomaly enhancement circuit and the background reconstruction circuit, and initialize parameters such as the number of superpixels, the size of the convolutional kernel, the learning rate, and the number of iterations.

[0060] In step S2, in the anomaly enhancement circuit, input the hyperspectral data into the anomaly detector to obtain an anomaly probability map, and use connected component analysis to screen the anomaly regions to generate an anomaly detection map. Adopt a moving average update strategy to obtain an anomaly probability detection map. Specifically:

[0061] Step S21, Neighborhood difference analysis: Input the hyperspectral data into the anomaly detector. The hyperspectral data is fused by band and superpixel segmented. Each pixel determines the degree of anomaly according to the difference from the surrounding pixels to obtain an anomaly probability map, and screen the anomaly regions from the consistency image through connected component analysis, and eliminate the regions with too large area to generate an anomaly detection map.

[0062] Step S22, Moving average update: Use the moving average update strategy for all anomaly detection maps to obtain an anomaly probability detection map.

[0063] Among them, the purpose of the anomaly detector is to detect a small number of abnormal pixels from the image, and these pixels are usually significantly different from the statistical characteristics of the surrounding areas. In hyperspectral data, anomaly detection not only needs to consider the differences in the spectral dimension, but also needs to combine the neighborhood information in the spatial dimension. For this reason, this embodiment proposes a hyperspectral anomaly detection method combining grayscale conversion, superpixel segmentation and neighborhood difference analysis. For each pixel , calculate its difference from the neighborhood pixels (from the superpixel region surrounding the current pixel), and the anomaly metric is measured by calculating the difference between the grayscale value of the current pixel and its neighborhood mean. Specifically:

[0064] Step S211, Sum the hyperspectral data by band to generate a grayscale image G, and retain the overall spectral information;

[0065] Step S212, Represent each pixel value of the grayscale image G as the sum of all bands:

[0066]

[0067] Normalize the grayscale image G to the interval [0, 1];

[0068] Step S213: Divide the normalized grayscale image G into multiple superpixel regions through a superpixel segmentation algorithm (such as the SLIC algorithm). The goal of superpixel segmentation is to group the pixels in the image so that each superpixel region contains similar pixel values, considering both the similarity of pixel grayscale values and spatial proximity. Assign a unique label to each superpixel , indicating the superpixel region to which the current pixel belongs;

[0069] Step S214: For each pixel , take the difference between the pixel value and the mean of its surrounding neighborhood as its degree of abnormality; First, obtain the current pixel value , and then extract the neighborhood pixel values from the superpixel it belongs to; Let the label of the current superpixel be , and set a distance parameter d, and find the pixels surrounding the superpixel through distance neighborhood expansion; Let the neighborhood pixel value be , that is, for each pixel at the pixel value, which belongs to the neighborhood of the superpixel within the distance d, and the formula for the neighborhood is:

[0070] = ;

[0071] Step S215: Calculate the mean difference between the current pixel value and the neighborhood pixel value :

[0072] ;

[0073] Step S216: In order to highlight the abnormal features, square the difference value further and record it in the anomaly probability map M, and perform normalization processing:

[0074] ;

[0075] Step S217: Since the anomalies account for a very small part of the whole, threshold screening can be carried out; Calculate the threshold T of the top 5% of the pixel values in :

[0076]

[0077] In the formula, Arrange all the pixel values in the anomaly probability map M in descending order, and convert the top 5% of the pixel values to 1;

[0078] Step S218: Perform connected region segmentation on the above pixels. Since the anomalies usually only account for a small part, in order to remove false anomalies, calculate the number of pixel points contained in each anomaly component for each anomaly component, and record each pixel block as , i = 1, 2, 3…, the threshold of the number of pixel points can be set by oneself. In this embodiment, we believe that if it is greater than 50, it is a pseudo - anomaly, that is, if the number of pixel points is greater than 50, then it is considered that this pixel block is not an anomaly point, denoted as 0; otherwise, the original value is retained.

[0079] Step S3, in the background reconstruction circuit, a global robust CNN is used to perform super - pixel partitioning on the hyperspectral data and construct a Laplacian matrix; the above operations are integrated into the convolutional neural network to reconstruct the hyperspectral data, and the error between the reconstructed data and the original data is used as an anomaly probability detection map;

[0080] In step S3, the convolutional neural network adopts an encoder - decoder network structure. Among them, the encoder network is used to extract local features, and after adding the Laplacian matrix to its output, it is input into the decoder network as the input of the decoder network. Specifically:

[0081] Step S31, represent the hyperspectral data received by the network as , where m is the width of the image, n is the height of the image, and b is the number of bands of the image;

[0082] Step S32, in the encoder network, input the hyperspectral data into the input layer. In the first convolutional layer, use 64 convolutional kernels of size 3×3×b for convolution operations, and at the same time adopt padding to keep the spatial dimension unchanged, so the output size is still m×n, and when the number of bands is reduced to 64;

[0083] Step S33, transfer the output data to the second convolutional layer. In this layer, use 32 convolutional kernels of size 3×3×64 for convolution, and also apply batch normalization and ReLU activation function. After this layer of processing, the dimension of the output data becomes m×n×32.

[0084] The above stage is mainly to complete the extraction and expression of local features. At the same time, in order to make up for the deficiency of the convolutional neural network in global feature modeling, the Laplacian matrix L is introduced, and the processed data is input into the decoder network. By combining with global features, the network further performs de - convolution operations on the data, restoring it to the original dimension m×n×b, realizing the reconstruction of hyperspectral data.

[0085] Considering that in practical applications, hyperspectral data usually has high spatial and spectral resolutions, directly processing the entire data cube may lead to excessive memory overhead and even inability to run. Therefore, this embodiment proposes a block - processing strategy, dividing the hyperspectral data into M×M blocks, and the data of each block is represented as Apply a convolutional neural network independently to each block. Through block processing, not only the memory requirement for a single calculation is reduced, but also the overall operation efficiency can be improved.

[0086] To accelerate the construction of the Laplacian matrix, a superpixel segmentation method (corresponding to the PCA&SLIC method in ①) is further introduced. The superpixel technology can divide the hyperspectral data into fewer pixel groups with consistent regions, obtaining a modified Laplacian matrix , thus significantly reducing the time complexity when calculating the Laplacian matrix. The objective function of the proposed convolutional neural network is as follows:

[0087] In Equation (1), is the reconstruction error, is the Laplacian matrix regularization term, which preserves the local geometric structure and spatial consistency, while encouraging samples within a locally homogeneous region to share similar representations. λ is the balance coefficient, N is the number of samples, that is, the total number of pixels, which is m×n×b, and Z is the latent representation matrix of the hidden layer.

[0088] Step S4, outlier determination: Multiply the outlier probability detection maps obtained from the outlier enhancement line and the background reconstruction line respectively to obtain the final outlier probability map;

[0089] Step S5, hyperspectral data update: Update the hyperspectral data input to the outlier enhancement line and the background reconstruction line in real time according to the final outlier probability map;

[0090] Step S6, detection map generation: After the iteration is completed, output the difference between the input data of the updated background reconstruction line and the original data as the outlier detection map.

[0091] For the outlier detection of hyperspectral data, by combining outlier enhancement and background reconstruction, the two complement and promote each other. As Figure 1 shown, in the initial state, represents the original hyperspectral data, where m is the width of the image, n is the height of the image, and b is the number of bands of the image. 、 respectively represent the inputs of the background reconstruction line and the outlier enhancement line at the k-th iteration, where , .

[0092] In the background reconstruction line (box ①), after reconstructing as the input, the error between the two is ;

[0093] ;

[0094] In the outlier enhancement line (box ②), Input it into an anomaly detector that combines superpixel technology and is based on comparison with surrounding pixel points, and obtain the enhanced anomaly component as , and at the same time adopt a moving average update strategy to enhance the robustness of the anomaly detector:

[0095] ;

[0096] Only when the anomaly detection maps obtained from both the background reconstruction circuit and the anomaly enhancement circuit consider this point as an anomaly point, this point is a real anomaly point. Therefore, finally, the anomaly weights of each pixel point are obtained as:

[0097] ;

[0098] In order to make the hyperspectral data obtained by the background reconstruction circuit more and more tend to the background, take as the input of the next background reconstruction circuit; at the same time, in order to facilitate easier enhancement of the anomaly part in the next iteration, take as the input of the next anomaly enhancement circuit. Since each iteration will cause the data of the anomaly points in to drop rapidly, so finally the difference between the hyperspectral data obtained from the last iteration and the original hyperspectral data can be output:

[0099] ;

[0100] Finally, perform spectral fusion on D along the channel dimension to obtain

[0101] ;

[0102] Finally, normalize D to obtain the final detection map D.

[0103] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A hyperspectral anomaly detection method based on a dual-line iterative convolutional neural network, characterized in that: The steps include: Step S1, input and parameter setting, input hyperspectral data, put the hyperspectral data into the anomaly enhancement circuit and background reconstruction circuit, initialize the number of superpixels, convolution kernel size, learning rate and number of iterations; Step S2: In the anomaly enhancement circuit, the hyperspectral data is input into the anomaly detector to obtain an anomaly probability map, and the connected region analysis is used to screen the abnormal region to generate an anomaly detection map, and the sliding average update strategy is adopted to obtain the anomaly probability detection map; Step S3, in the background reconstruction circuit, the global robust CNN divides the hyperspectral data into superpixels and constructs a Laplacian matrix; the above operations are integrated into the convolutional neural network to reconstruct the hyperspectral data, and the error between the reconstructed data and the original data is used as an abnormal probability detection map; Step S4, abnormal point determination, performing point multiplication on the abnormal probability detection maps obtained by the abnormal enhancement line and the background reconstruction line respectively to obtain the final abnormal probability map; Step S5, updating the hyperspectral data, updating the hyperspectral data input by the anomaly enhancement line and the background reconstruction line in real time according to the final anomaly probability map; Step S6, generating a detection map. After the iteration is completed, the difference between the input data of the updated background reconstruction line and the original data is output as an anomaly detection map; In step S3, the convolutional neural network adopts an encoding-decoding network structure. The encoding network is used to extract local features, and the Laplacian matrix is ​​added to its output before being input into the decoding network as the input of the decoding network. Step S2 specifically includes the following steps: Step S21, domain difference analysis, the hyperspectral data is put into the anomaly detector, the hyperspectral data is fused by band and divided into superpixels, the degree of abnormality of each pixel is judged according to the degree of difference with the surrounding pixels, and an abnormal probability map is obtained. The abnormal area is screened from the consistency image through connected region analysis, and the area with too large area is eliminated to generate an anomaly detection map; Step S22, sliding average update, all anomaly detection maps are updated using the sliding average update strategy to obtain anomaly probability detection maps; In step S2, the detection of the anomaly detector includes the following steps: Step S211: Hyperspectral data Sum by band to generate grayscale image G; Step S212, each pixel value of the grayscale image G is expressed as the sum of all bands: Normalize the grayscale image G to the interval [0, 1]; where m is the width of the image, n is the height of the image, and b is the number of bands of the image. is the grayscale image G The value of the pixel; Step S213: divide the normalized grayscale image G into multiple superpixel regions using a superpixel segmentation algorithm, and assign a unique label to each superpixel. , represents the superpixel area to which the current pixel belongs; Step S214, set the label of the current superpixel to , and set a distance parameter d, find the pixels surrounding the superpixel by distance field expansion, and set the current pixel value to , the pixel value of the area is , the calculation formula of the field is: = ； Step S215, calculate the current pixel value With the field pixel value The mean difference of : ； Step S216, square the difference value and record it in the abnormal probability map M, and perform standardization processing: ； Step S217, calculate The threshold T of the top 5% of pixel values ​​is: In the formula, Arrange all pixel values ​​in the abnormal probability map M in descending order and convert the first 5% of pixel values ​​to 1; Step S218, perform connected region segmentation, calculate the number of pixels contained in each abnormal component, and record each pixel block as , i=1,2,3…; if If the number of pixels is greater than 50, the pixel block is considered not to be an outlier and is recorded as 0; otherwise, the original value is retained.

2. According to claim 1, a hyperspectral anomaly detection method based on dual-line iterative convolutional neural network is characterized in that: The encoding network extracts local features including the following steps: Step S31, the original hyperspectral data received by the network is represented as , m is the width of the image, n is the height of the image, and b is the number of bands of the image; Step S32, in the encoding network, the hyperspectral data is input into the input layer, and in the first convolution layer, 64 convolution kernels of size 3×3×b are used to perform convolution operation, and the data is output after padding; In step S33, the output data is passed to the second convolutional layer, and convolution is performed using 32 convolution kernels of size 3×3×64. The data is output after batch normalization and ReLU activation function.

3. According to claim 2, a method for detecting anomalies of a hyperspectral spectrum based on a dual-line iterative convolutional neural network is characterized in that: The hyperspectral data is divided into M×M blocks, and the data of each block is represented as , and apply convolutional neural networks independently on each block.

4. According to claim 1, a method for hyperspectral anomaly detection based on dual-line iterative convolutional neural network is characterized in that: In step S3, the hyperspectral data is divided into pixel groups with consistent regions by introducing a superpixel segmentation algorithm, and the modified Laplace matrix is ​​obtained. , using the modified Laplace matrix To speed up the construction of the Laplace matrix, the objective function of the proposed convolutional neural network is as follows: (1) In formula (1), is the reconstruction error, is the Laplace matrix regularization term, λ is the balance coefficient, N is the number of samples, Z is the potential representation matrix of the hidden layer, is the original hyperspectral data, It is the reconstructed data of the original spectral data obtained after iterative processing.

Images