Hyperspectral image anomaly detection method and device based on nested autoencoder

Abstract

The present invention relates to the field of image anomaly detection technology, and in particular to a method and device for detecting anomalies in hyperspectral images based on nested autoencoders, wherein the method comprises: obtaining an original hyperspectral image; based on an upsampling structure, using a first-level encoder to perform feature encoding and decoding on the original hyperspectral image in a spatial dimension, and using a second-level encoder to perform feature encoding and decoding on the original hyperspectral image in a spectral dimension, to obtain spatial features and spectral features of the original hyperspectral image; and reconstructing abnormal pixel points of the reconstructed hyperspectral image based on the spatial features and spectral features. Thus, the problems of high false alarm rate, poor detail reconstruction effect, and low image information utilization rate of the prior art detection method are solved, and by improving the structure of the deep learning model, the method has the advantages of automation, high detection accuracy, low complexity, fast and high efficiency, etc.

Description

Technical Field

[0001] The present invention relates to the technical field of image anomaly detection, and in particular to a hyperspectral image anomaly detection method and device based on a nested autoencoder. Background Art

[0002] In traditional deep learning-based anomaly detection methods, convolutional autoencoders are mostly used to reconstruct the background. However, it can only mine the differences between the anomaly and the background in the spatial dimension, ignoring the abnormal features that are significantly different from the background in the spectral dimension, making the reconstructed background easily disturbed by abnormal pixels, resulting in a high false alarm rate. Traditional convolutional autoencoders only have a downsampling structure in the spatial dimension and cannot effectively reconstruct certain details. Using only the reconstruction error in the spatial dimension for anomaly detection is not enough to utilize the spectral information of hyperspectral images. Summary of the invention

[0003] The present invention provides a hyperspectral image anomaly detection method and device based on a nested autoencoder to solve the problems of high false alarm rate, poor detail reconstruction effect, low image information utilization rate, etc. in the prior art detection method.

[0004] The first aspect of the present invention provides a method for detecting anomalies in a hyperspectral image based on a nested autoencoder. The method adopts an improved deep learning model, and the improved deep learning model includes an upsampling structure and a downsampling structure. The downsampling structure includes a first-level encoder and a second-level encoder. A convolutional layer with a stride of 2 is used to replace the pooling layer of the deep learning model. The method includes the following steps: obtaining an original hyperspectral image; based on the upsampling structure, using the first-level encoder to perform feature encoding and decoding on the original hyperspectral image in a spatial dimension, and using the second-level encoder to perform feature encoding and decoding on the original hyperspectral image in a spectral dimension to obtain spatial features and spectral features of the original hyperspectral image; reconstructing abnormal pixel points of a reconstructed hyperspectral image based on the spatial features and the spectral features.

[0005] Optionally, after obtaining the abnormal pixel points of the reconstructed hyperspectral image based on the spatial features and the spectral features, the method includes: calculating the spatial reconstruction error at each pixel position of the original hyperspectral image and the reconstructed hyperspectral image and the spectral reconstruction error of each pixel in different spectral bands; adjusting the weight of the spatial reconstruction error and the weight of the spectral reconstruction error based on a preset adjustment strategy, and optimizing the reconstruction capability of the deep learning model based on the adjusted weight of the spatial reconstruction error and the weight of the spectral reconstruction error.

[0006] Optionally, the calculation formula of the spatial reconstruction error is:

[0007] ;

[0008] in, is the image height, is the image width, is the number of spectral bands, =1... , is the original hyperspectral image, To reconstruct hyperspectral images, is the pixel position.

[0009] Optionally, the calculation formula of the spectrum reconstruction error is:

[0010] .

[0011] The second aspect of the present invention provides a hyperspectral image anomaly detection device based on a nested autoencoder, wherein the device adopts an improved deep learning model, wherein the improved deep learning model includes an upsampling structure and a downsampling structure, wherein the downsampling structure includes a first-level encoder and a second-level encoder, and a convolution layer with a stride of 2 is used to replace the pooling layer of the deep learning model. The device includes: a first acquisition module, used to acquire an original hyperspectral image; a second acquisition module, used to perform feature encoding and decoding on the original hyperspectral image in a spatial dimension using the first-level encoder based on the upsampling structure, and perform feature encoding and decoding on the original hyperspectral image in a spectral dimension using the second-level encoder to obtain spatial features and spectral features of the original hyperspectral image; and a detection module, used to reconstruct abnormal pixel points of a reconstructed hyperspectral image based on the spatial features and the spectral features.

[0012] Optionally, after obtaining abnormal pixel points of the reconstructed hyperspectral image based on the spatial features and the spectral features, the detection module is further used to: calculate the spatial reconstruction error at each pixel position of the original hyperspectral image and the reconstructed hyperspectral image and the spectral reconstruction error of each pixel in different spectral bands; adjust the weight of the spatial reconstruction error and the weight of the spectral reconstruction error based on a preset adjustment strategy, and optimize the reconstruction capability of the deep learning model based on the adjusted weight of the spatial reconstruction error and the weight of the spectral reconstruction error.

[0013] Optionally, the calculation formula of the spatial reconstruction error is:

[0014] ;

[0015] in, is the image height, is the image width, is the number of spectral bands, =1... , is the original hyperspectral image, To reconstruct hyperspectral images, is the pixel position.

[0016] Optionally, the calculation formula of the spectrum reconstruction error is:

[0017] .

[0018] A third aspect of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the hyperspectral image anomaly detection method based on a nested autoencoder as described in the above embodiment.

[0019] A fourth aspect of the present invention provides a computer program product having a computer program stored thereon, which is executed by a processor to implement the hyperspectral image anomaly detection method based on a nested autoencoder as described in the above embodiment.

[0020] In the above implementation, the original hyperspectral image is obtained, and based on the upsampling structure, the first-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spatial dimension, and the second-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spectral dimension, so as to obtain the spatial features and spectral features of the original hyperspectral image, and the abnormal pixel points of the reconstructed hyperspectral image are reconstructed based on the spatial features and spectral features. Thus, the problems of high false alarm rate, poor detail reconstruction effect, and low image information utilization rate of the prior art detection method are solved, and by improving the structure of the deep learning model, it has the advantages of automation, high detection accuracy, low complexity, fast and efficient.

[0021] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0023] Figure 1 A flowchart of a method for anomaly detection in a hyperspectral image based on a nested autoencoder according to an embodiment of the present invention;

[0024] Figure 2 Flow chart of a method for anomaly detection in hyperspectral images based on nested autoencoders according to an embodiment of the present invention;

[0025] Figure 3 is an example diagram of a hyperspectral image anomaly detection device based on a nested autoencoder according to an embodiment of the present invention;

[0026] Figure 4 FIG. 4 is a schematic diagram of a structure of an electronic device according to an embodiment of the present invention. Description of the drawings:

[0028] 10-hyperspectral image anomaly detection device based on nested autoencoder; 100-first acquisition module; 200-second acquisition module; 300-detection module; 401-memory; 402-processor; 403-communication interface. DETAILED DESCRIPTION

[0029] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0030] The following describes the hyperspectral image anomaly detection method and device based on nested autoencoders of the embodiment of the present invention with reference to the accompanying drawings. In view of the problems of high false alarm rate, poor detail reconstruction effect and low image information utilization rate of the prior art detection method mentioned in the above background technology, the present invention provides a hyperspectral image anomaly detection method based on nested autoencoders, in which the original hyperspectral image is obtained, and based on the upsampling structure, the first-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spatial dimension, and the second-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spectral dimension to obtain the spatial features and spectral features of the original hyperspectral image, and the abnormal pixel points of the reconstructed hyperspectral image are reconstructed based on the spatial features and spectral features. Thus, the problems of high false alarm rate, poor detail reconstruction effect and low image information utilization rate of the prior art detection method are solved, and by improving the structure of the deep learning model, it has the advantages of automation, high detection accuracy, low complexity, fast and efficient.

[0031] Specifically, Figure 1 A schematic diagram of a process for detecting anomalies in hyperspectral images based on a nested autoencoder provided in an embodiment of the present invention.

[0032] The method adopts an improved deep learning model, which includes an upsampling structure and a downsampling structure. The downsampling structure includes a first-level encoder and a second-level encoder. The convolution layer with a stride of 2 is used to replace the pooling layer of the deep learning model.

[0033] Specific as Figure 2As shown, an embodiment of the present invention improves the downsampling structure of the deep learning model, specifically replacing the pooling layer with a convolution with a stride of 2 to extract spatial features.

[0034] Since the pooling layer of the convolutional neural network has no learnable parameters, it can only perform fixed operations and cannot optimize the feature extraction process through learning. In addition, the pooling layer will lose some information within the pooling window, especially the maximum pooling, which only retains the maximum value within the window, causing other information to be ignored. The convolution layer has learnable parameters (convolution kernels) that can be optimized through back propagation, and can extract multiple features through different convolution kernels, not limited to local extreme values ​​or means, and can capture richer patterns and details. In addition, the convolution operation with a stride of 2 can halve the width and height of the feature map to achieve effective downsampling while retaining important spatial features. The 3*3 convolution kernel can effectively capture local spatial features and has a lower computational cost than larger convolution kernels (such as 5*5 or 7*7).

[0035] Specific as Figure 1 As shown, the hyperspectral image anomaly detection method based on nested autoencoder includes the following steps:

[0036] In step S101, an original hyperspectral image is acquired.

[0037] In step S102, based on the upsampling structure, the first-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spatial dimension, and the second-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spectral dimension to obtain the spatial features and spectral features of the original hyperspectral image.

[0038] In an embodiment of the present invention, an upsampling structure is added to the deep learning model, and a convolution interpolation method is used to reconstruct spatial features.

[0039] Among them, the upsampling structure uses the convolution interpolation method to reconstruct the spatial features. The interpolation method uses bilinear interpolation. Assuming that the 4 data points are known to be , , , , then the interpolation point The values ​​are:

[0040]

[0041] The convolution interpolation method can increase the spatial size of the feature map and upsample the low-resolution feature map to a higher resolution. After interpolation, the generated feature map may contain some artifacts or non-smooth areas. Applying a convolution operation to the interpolated feature map can further optimize the interpolated feature map and enhance its detail information.

[0042] In step S103, abnormal pixel points of a reconstructed hyperspectral image are reconstructed based on the spatial features and the spectral features.

[0043] Specifically, a two-stage autoencoder is nested, including a first-stage encoder and a second-stage encoder. The first-stage encoder is a spatial dimension autoencoder, and the second-stage encoder is a spectral dimension autoencoder to simultaneously capture the spatial and spectral features in the hyperspectral image. The first-stage encoder mainly processes the feature encoding and decoding of the spatial dimension, while the second-stage encoder performs encoding and decoding on the spectral dimension.

[0044] Specifically, assuming there is an original hyperspectral image , the shape is , input it into the spatial dimension encoder to obtain the spatial feature, indicating , the shape is ,Will Input to the spectral dimension encoder to obtain the spectral features, indicating , the shape is The above process completes the encoding of the original hyperspectral image. Input into the spectral dimension decoder to obtain the spectral features after spectral decoding , the shape is ,Will Input to the spatial dimension decoder to obtain the reconstructed hyperspectral image , the shape is The above process completes the decoding of the feature image, reconstructs the hyperspectral image, and can obtain the abnormal pixel points of the reconstructed hyperspectral image.

[0045] Therefore, through the nested structure of the two-stage autoencoder, the complex features in the hyperspectral image can be better extracted and reconstructed, thereby improving the effect of anomaly detection.

[0046] Optionally, in some embodiments, after obtaining abnormal pixel points of the reconstructed hyperspectral image based on spatial features and spectral features, the method includes: calculating the spatial reconstruction error at each pixel position of the original hyperspectral image and the reconstructed hyperspectral image and the spectral reconstruction error of each pixel in different spectral bands; adjusting the weight of the spatial reconstruction error and the weight of the spectral reconstruction error based on a preset adjustment strategy, and optimizing the reconstruction capability of the deep learning model based on the adjusted weight of the spatial reconstruction error and the weight of the spectral reconstruction error.

[0047] Optionally, in some embodiments, the calculation formula of the spatial reconstruction error is:

[0048] ;

[0049] in, is the image height, is the image width, is the number of spectral bands, =1... , is the original hyperspectral image, To reconstruct hyperspectral images, is the pixel position.

[0050] Optionally, in some embodiments, the calculation formula of the spectrum reconstruction error is:

[0051] .

[0052] Specifically, according to the preset adjustment strategy, the weights of the spatial reconstruction error and the spectral reconstruction error are dynamically adjusted according to the current number of training rounds of the deep learning model. The reconstructed hyperspectral image is the output data generated by the deep learning model after passing through the encoder and decoder. The reconstruction error reflects the similarity between the original data and the reconstructed data. The smaller the error, the better the model reconstruction effect. Compared with ordinary images, hyperspectral images have increased spectral information, so the model reconstruction effect of hyperspectral images can be quantified by spatial reconstruction error and spectral reconstruction error.

[0053] Specifically, assuming that the input original hyperspectral image X and the reconstructed hyperspectral image The dimensions of , where H is the image height, W is the image width, and C is the number of spectral bands.

[0054] The spatial reconstruction error can be realized by calculating the mean square error at each pixel position. , calculate its reconstruction error in all spectral bands and then take the average; the spectral reconstruction error can be achieved by calculating the mean square error of each pixel in different spectral bands.

[0055] Since the spatial reconstruction error is mainly focused on in the early stage of training to let the model learn to reconstruct the input data, and the spectral reconstruction error is gradually increased in the later stage of training to improve the quality of model reconstruction. Therefore, the weight of the spatial reconstruction error should decrease with the increase of training rounds, and the weight of the spectral reconstruction error should increase with the increase of training rounds. The weight update is realized through the callback function. The callback function updates the current training round at the end of each training round, so that the reconstruction ability of the deep learning model is optimized according to the adjusted weight of the spatial reconstruction error and the weight of the spectral reconstruction error.

[0056] According to the hyperspectral image anomaly detection method based on nested autoencoders proposed in an embodiment of the present invention, the original hyperspectral image is obtained, and based on the upsampling structure, the first-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spatial dimension, and the second-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spectral dimension, so as to obtain the spatial features and spectral features of the original hyperspectral image, and the abnormal pixel points of the reconstructed hyperspectral image are reconstructed based on the spatial features and spectral features. Thus, the problems of high false alarm rate, poor detail reconstruction effect, low image information utilization rate, etc. of the prior art detection method are solved, and by improving the structure of the deep learning model, it has the advantages of automation, high detection accuracy, low complexity, fast and efficient.

[0057] Next, a hyperspectral image anomaly detection device based on a nested autoencoder proposed in an embodiment of the present invention is described with reference to the accompanying drawings.

[0058] Figure 3 4 is a block diagram of a hyperspectral image anomaly detection device based on a nested autoencoder according to an embodiment of the present invention.

[0059] The device adopts an improved deep learning model, which includes an upsampling structure and a downsampling structure. The downsampling structure includes a first-level encoder and a second-level encoder. A convolution layer with a stride of 2 is used to replace the pooling layer of the deep learning model. The hyperspectral image anomaly detection device 10 based on nested autoencoders includes: a first acquisition module 100, a second acquisition module 200 and a detection module 300.

[0060] Among them, the first acquisition module 100 is used to acquire the original hyperspectral image; the second acquisition module 200 is used to perform feature encoding and decoding on the original hyperspectral image in the spatial dimension using the first-level encoder based on the upsampling structure, and perform feature encoding and decoding on the original hyperspectral image in the spectral dimension using the second-level encoder to obtain the spatial features and spectral features of the original hyperspectral image; the detection module 300 is used to reconstruct abnormal pixel points of the reconstructed hyperspectral image based on the spatial features and spectral features.

[0061] Optionally, in some embodiments, after obtaining abnormal pixel points of the reconstructed hyperspectral image based on spatial features and spectral features, the detection module 300 is further used to: calculate the spatial reconstruction error at each pixel position of the original hyperspectral image and the reconstructed hyperspectral image and the spectral reconstruction error of each pixel in different spectral bands; adjust the weight of the spatial reconstruction error and the weight of the spectral reconstruction error based on a preset adjustment strategy, and optimize the reconstruction capability of the deep learning model based on the adjusted weight of the spatial reconstruction error and the weight of the spectral reconstruction error.

[0062] Optionally, in some embodiments, the calculation formula of the spatial reconstruction error is:

[0063] ;

[0064] in, is the image height, is the image width, is the number of spectral bands, =1... , is the original hyperspectral image, To reconstruct hyperspectral images, is the pixel position.

[0065] Optionally, in some embodiments, the calculation formula of the spectrum reconstruction error is:

[0066] .

[0067] It should be noted that the aforementioned explanation of the embodiment of the hyperspectral image anomaly detection method based on nested autoencoders is also applicable to the hyperspectral image anomaly detection device based on nested autoencoders of this embodiment, and will not be repeated here.

[0068] According to the hyperspectral image anomaly detection device based on nested autoencoders proposed in the embodiment of the present invention, the original hyperspectral image is obtained, and based on the upsampling structure, the first-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spatial dimension, and the second-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spectral dimension, so as to obtain the spatial features and spectral features of the original hyperspectral image, and the abnormal pixel points of the reconstructed hyperspectral image are reconstructed based on the spatial features and spectral features. Thus, the problems of high false alarm rate, poor detail reconstruction effect, low image information utilization rate, etc. of the prior art detection method are solved, and by improving the structure of the deep learning model, it has the advantages of automation, high detection accuracy, low complexity, fast and efficient.

[0069] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device may include:

[0070] Memory 401 , processor 402 , and a computer program stored in the memory 401 and executable on the processor 402 .

[0071] When the processor 402 executes the program, the hyperspectral image anomaly detection method based on the nested autoencoder provided in the above embodiment is implemented.

[0072] Furthermore, the electronic device further comprises:

[0073] The communication interface 403 is used for communication between the memory 401 and the processor 402 .

[0074] The memory 401 is used to store computer programs that can be executed on the processor 402 .

[0075] The memory 401 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0076] If the memory 401, the processor 402 and the communication interface 403 are implemented independently, the communication interface 403, the memory 401 and the processor 402 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0077] Optionally, in a specific implementation, if the memory 401, the processor 402 and the communication interface 403 are integrated on a chip, the memory 401, the processor 402 and the communication interface 403 can communicate with each other through an internal interface.

[0078] The processor 402 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0079] The embodiment of the present invention further provides a computer program product, on which a computer program is stored, and when the program is executed by a processor, the above rainfall detection method is implemented.

[0080] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0081] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "N" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0082] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present invention belong.

[0083] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be specifically implemented in any computer program product for use with an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, a "computer program product" can be any device that can contain, store, communicate, propagate or transmit a program for use with an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer program products (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). Furthermore, the computer program product may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example by optically scanning the paper or other medium and then editing, interpreting or, if necessary, processing in another suitable manner, and then storing it in a computer memory.

[0084] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0085] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer program product, which, when executed, includes one or a combination of the steps of the method embodiment.

[0086] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer program product.

[0087] The computer program product mentioned above may be a read-only memory, a disk or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A hyperspectral image anomaly detection method based on nested autoencoders, characterized in that: The method adopts an improved deep learning model, the improved deep learning model includes an upsampling structure and a downsampling structure, the downsampling structure includes a first-stage encoder and a second-stage encoder, and a convolutional layer with a stride of 2 is used to replace the pooling layer of the deep learning model. The method includes the following steps: Obtaining original hyperspectral images; Based on the upsampling structure, the first-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spatial dimension, and the second-level encoder is used to perform feature encoding and decoding on the original hyperspectral image in the spectral dimension to obtain the spatial features and spectral features of the original hyperspectral image; Reconstruct abnormal pixel points of a reconstructed hyperspectral image based on the spatial features and the spectral features; After the abnormal pixel points of the reconstructed hyperspectral image are reconstructed based on the spatial features and the spectral features, the method includes: calculating the spatial reconstruction error at each pixel position of the original hyperspectral image and the reconstructed hyperspectral image and the spectral reconstruction error of each pixel in different spectral bands; adjusting the weight of the spatial reconstruction error and the weight of the spectral reconstruction error based on a preset adjustment strategy, and optimizing the reconstruction capability of the deep learning model based on the adjusted weight of the spatial reconstruction error and the weight of the spectral reconstruction error; The calculation formula of the spatial reconstruction error is: ； in, is the image height, is the image width, is the number of spectral bands, =1... , is the original hyperspectral image, To reconstruct the hyperspectral image, is the pixel position.

2. The method for anomaly detection of hyperspectral images based on nested autoencoders according to claim 1, characterized in that: The calculation formula of the spectrum reconstruction error is: 。 3. A hyperspectral image anomaly detection device based on nested autoencoders, characterized in that: The device adopts an improved deep learning model, the improved deep learning model includes an upsampling structure and a downsampling structure, the downsampling structure includes a first-stage encoder and a second-stage encoder, and a convolution layer with a stride of 2 is used to replace the pooling layer of the deep learning model. The device includes: A first acquisition module is used to acquire an original hyperspectral image; A second acquisition module is used to perform feature encoding and decoding on the original hyperspectral image in the spatial dimension using the first-level encoder based on the upsampling structure, and perform feature encoding and decoding on the original hyperspectral image in the spectral dimension using the second-level encoder to obtain the spatial features and spectral features of the original hyperspectral image; A detection module, used for reconstructing abnormal pixel points of a reconstructed hyperspectral image based on the spatial features and the spectral features; After the abnormal pixel points of the reconstructed hyperspectral image are reconstructed based on the spatial features and the spectral features, the detection module is further used to: calculate the spatial reconstruction error at each pixel position of the original hyperspectral image and the reconstructed hyperspectral image and the spectral reconstruction error of each pixel in different spectral bands; adjust the weight of the spatial reconstruction error and the weight of the spectral reconstruction error based on a preset adjustment strategy, and optimize the reconstruction capability of the deep learning model based on the adjusted weight of the spatial reconstruction error and the weight of the spectral reconstruction error; The calculation formula of the spatial reconstruction error is: ； in, is the image height, is the image width, is the number of spectral bands, =1... , is the original hyperspectral image, To reconstruct the hyperspectral image, is the pixel position.

4. The hyperspectral image anomaly detection device based on nested autoencoder according to claim 3, characterized in that: The calculation formula of the spectrum reconstruction error is: 。 5. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the hyperspectral image anomaly detection method based on a nested autoencoder as claimed in any one of claims 1 to 2.

6. A computer program product, wherein the computer program product stores a computer program, characterized in that: When the program is executed by a processor, the method for detecting anomalies in a hyperspectral image based on a nested autoencoder as described in any one of claims 1 to 2 is implemented.

Images