Hyperspectral image target detection method and device based on deep learning, storage medium and equipment

Abstract

The embodiment of the invention relates to a hyperspectral image target detection method and device based on deep learning, a storage medium and equipment. The method comprises the following steps: acquiring a preprocessed spectral remote sensing image; processing the preprocessed spectral remote sensing image based on a pre-constructed three-dimensional convolutional neural network to obtain a target image after classification detection; wherein the three-dimensional convolutional neural network comprises a first three-dimensional convolutional layer, an activation layer, a second three-dimensional convolutional layer, an activation layer and a full-connection layer which are connected in sequence, outputting a target image after classification detection, extracting target features by adopting the two convolutional layers, realizing feature extraction while considering the model training speed, and improving the accuracy of feature extraction. According to the method, the model can quickly and accurately obtain a training result, the ReLU function is adopted as the activation function, and compared with other activation functions, the ReLU function is simple in calculation and more efficient, and meanwhile, the problem of gradient disappearance is effectively avoided through a specific positive value output mode of the ReLU function.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of image detection technology, and in particular to a method, apparatus, storage medium and device for hyperspectral image target detection based on deep learning. Background Art

[0002] Spectral remote sensing imaging technology has been successfully applied to the aerospace field as a new round of detection and reconnaissance means. Airborne spectral remote sensing images have the characteristics of high resolution and rich spectral information bands, and play an important role in target detection and interpretation. Traditional images generally only have two spatial dimensions of information, while spectral images have the characteristics of "image and spectrum integration". In addition to spatial information, their images also contain spectral information, which has important potential in target detection and interpretation. As a key data source, spectral remote sensing images contain information about the material of objects in the spectral dimension, which provides a new perspective for target detection, making the position of image spatial information in spectral image target detection relatively weakened.

[0003] There have been many studies on intelligent assisted recognition methods for spectral remote sensing images at home and abroad. At present, the research bottleneck of intelligent assisted recognition methods for spectral remote sensing images lies in the complex imaging mechanism of spectral remote sensing images, huge amount of data and redundancy, slow algorithm model training, low accuracy, differences in imaging areas of different bands, difficulty in registration, affecting classification accuracy and poor algorithm generalization ability. Summary of the invention

[0004] In view of this, the embodiments of the present application propose a hyperspectral image target detection method, device, storage medium and equipment based on deep learning, which aims to extract target features by using the two convolutional layers of the 3D-CNN architecture, while ensuring the extraction of features and taking into account the model training speed, so that image detection results can be obtained quickly and accurately.

[0005] To achieve the above-mentioned purpose, an embodiment of the present application provides a hyperspectral image target detection method based on deep learning, comprising: obtaining a preprocessed spectral remote sensing image; processing the preprocessed spectral remote sensing image based on a prebuilt three-dimensional convolutional neural network to obtain a target image after classification detection; wherein the three-dimensional convolutional neural network comprises a first three-dimensional convolutional layer, an activation layer, a second three-dimensional convolutional layer, an activation layer and a fully connected layer connected in sequence, and the convolution kernel sizes of the first three-dimensional convolutional layer and the second three-dimensional convolutional layer are both 3×3×7, that is, the spatial dimension of the convolution kernel is 3, the spectral dimension of the convolution kernel is 7, the number of input nodes of the fully connected layer is 64, and the number of output nodes is 1; and the target image after classification detection is output.

[0006] Optionally, before the spectral remote sensing image is processed based on the pre-built three-dimensional convolutional neural network to obtain the target image after classification detection, the hyperspectral image target detection method based on deep learning also includes: using an atmospheric correction algorithm and a geometric correction algorithm to process the Pavia University data set to obtain a preprocessed data set; using the preprocessed data set to train multiple three-dimensional convolutional neural networks, and obtaining different overall accuracy values ​​by changing the parameters of the multiple three-dimensional convolutional neural networks, wherein the parameters of the three-dimensional convolutional neural network include the number of iterations, the convolution kernel size, and the random discard value; and selecting the three-dimensional convolutional neural network corresponding to the maximum overall accuracy value as the three-dimensional convolutional neural network to be evaluated.

[0007] Optionally, after selecting the three-dimensional convolutional neural network corresponding to the maximum overall accuracy value as the three-dimensional convolutional neural network to be evaluated, the deep learning-based hyperspectral image target detection method further includes: evaluating the three-dimensional convolutional neural network to be evaluated based on the confusion matrix of the spectral remote sensing image to obtain the pre-constructed three-dimensional convolutional neural network.

[0008] Optionally, before processing the preprocessed spectral remote sensing image based on the preconstructed three-dimensional convolutional neural network, processing the preprocessed spectral remote sensing image based on the preconstructed three-dimensional convolutional neural network also includes: acquiring a spectral remote sensing image; processing the spectral remote sensing image based on a radiation calibration algorithm, a histogram equalization algorithm and a median filter to obtain the preprocessed spectral remote sensing image.

[0009] To achieve the above-mentioned purpose, an embodiment of the present application also provides a hyperspectral image target detection device based on deep learning, characterized in that it includes: an acquisition module for acquiring a preprocessed spectral remote sensing image; a processing module for processing the preprocessed spectral remote sensing image based on a prebuilt three-dimensional convolutional neural network to obtain a target image after classification detection; wherein the three-dimensional convolutional neural network includes a first three-dimensional convolutional layer, an activation layer, a second three-dimensional convolutional layer, an activation layer and a fully connected layer connected in sequence, and the convolution kernel sizes of the first three-dimensional convolutional layer and the second three-dimensional convolutional layer are both 337, that is, the spatial dimension of the convolution kernel is 3, the spectral dimension of the convolution kernel is 7, the number of input nodes of the fully connected layer is 64, and the number of output nodes is 1; an output module is used to output the target image after classification detection.

[0010] To achieve the above-mentioned objectives, an embodiment of the present application also provides a terminal / electronic device / server, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any of the above-mentioned embodiments to provide a hyperspectral image target detection method based on deep learning.

[0011] To achieve the above objectives, an embodiment of the present application further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the hyperspectral image target detection method based on deep learning provided in any of the above embodiments can be implemented.

[0012] To achieve the above objectives, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the above-mentioned hyperspectral image target detection method based on deep learning.

[0013] The embodiments of the present application propose a method, device, storage medium and equipment for hyperspectral image target detection based on deep learning, which obtains a preprocessed spectral remote sensing image; processes the preprocessed spectral remote sensing image based on a pre-constructed three-dimensional convolutional neural network to obtain a target image after classification detection; wherein the three-dimensional convolutional neural network includes a first three-dimensional convolutional layer, an activation layer, a second three-dimensional convolutional layer, an activation layer and a fully connected layer connected in sequence, and the convolution kernel sizes of the first three-dimensional convolutional layer and the second three-dimensional convolutional layer are both 3×3×7, that is, the spatial dimension of the convolution kernel is 3, the spectral dimension of the convolution kernel is 7, the number of input nodes of the fully connected layer is 64, and the number of output nodes is 1; the target image after classification detection is output, and target features are extracted by using two convolutional layers to achieve the extraction of features while taking into account the model training speed, so that the model can quickly and accurately obtain training results, and the activation function uses the ReLU function. Compared with other activation functions, the ReLU function is simple to calculate, making the ReLU function more efficient. At the same time, the positive output mode unique to the ReLU function effectively avoids the gradient disappearance problem. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flow chart of a hyperspectral image target detection method based on deep learning provided in one embodiment of the present application;

[0015] Figure 2 It is a false color image of the Pavia University dataset of a hyperspectral image target detection method based on deep learning provided in one embodiment of the present application;

[0016] Figure 3 is an experimental result diagram of a hyperspectral image target detection method based on deep learning provided in one embodiment of the present application;

[0017] Figure 4 It is a schematic diagram of the structure of a hyperspectral image target detection device based on deep learning provided in one embodiment of the present application;

[0018] Figure 5 It is a schematic diagram of the model structure of a hyperspectral image target detection method based on deep learning provided in one embodiment of the present application. DETAILED DESCRIPTION

[0019] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is for the convenience of description, and the specific implementation of the present application should not be construed as any limitation, and the various embodiments can be combined and referenced with each other under the premise of no contradiction.

[0020] How to find effective features is the core problem of image classification, detection and recognition. In the past few decades, scientists have carried out a large number of research projects and found that biological visual systems can solve classification, detection and recognition problems very effectively. Machine learning researchers drew inspiration from this and designed a convolutional neural network model (3D-CNN). This neural network designed based on the response mechanism of neurons to specific areas in the visual field can capture the local features of the image, and gradually abstract and integrate these features through a hierarchical network structure, greatly reducing the number of parameters of the model and improving the calculation speed and accuracy. Because of the unique advantages of 3D-CNN for image processing, the present invention selects 3D-CNN as a method for spectral assisted recognition.

[0021] An embodiment of the present application proposes a hyperspectral image target detection method based on deep learning, which is applied to an electronic device, wherein the electronic device can be a terminal or a server. This embodiment and the following embodiments are all described by taking the server as an example. The implementation details of the hyperspectral image target detection method based on deep learning proposed in this embodiment are described in detail below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this solution.

[0022] The specific process of the hyperspectral image target detection method based on deep learning proposed in this embodiment can be as follows: Figure 1As shown in Figure 2, the hyperspectral image target detection method based on deep learning includes:

[0023] S10, obtaining a preprocessed spectral remote sensing image;

[0024] In an embodiment of the present invention, before step S10, the hyperspectral image target detection method based on deep learning may further include the following execution steps:

[0025] S01. Acquire spectral remote sensing images;

[0026] S02. Processing the spectral remote sensing image based on a radiation calibration algorithm, histogram equalization and median filtering to obtain a preprocessed spectral remote sensing image.

[0027] Among them, the radiation calibration algorithm is used to convert the digital quantization value (DN) captured by the sensor into a radiation brightness value with physical meaning, so as to eliminate the error of the sensor itself and determine the accurate radiation value at the entrance pupil of the sensor. The histogram equalization algorithm is used to enhance the image contrast and make the histogram distribution of the image more uniform. The median filter is used to remove salt and pepper noise in the image.

[0028] S20. Processing the preprocessed spectral remote sensing image based on a pre-built three-dimensional convolutional neural network to obtain a target image after classification detection; wherein the three-dimensional convolutional neural network includes a first three-dimensional convolutional layer, a first activation layer, a second three-dimensional convolutional layer, a second activation layer and a fully connected layer connected in sequence, and the convolution kernel sizes of the first three-dimensional convolutional layer and the second three-dimensional convolutional layer are both 3×3×7, that is, the spatial dimension of the convolution kernel is 3, the spectral dimension of the convolution kernel is 7, the number of input nodes of the fully connected layer is 64, the number of output nodes is 1, and the first activation layer and the second activation layer both use the ReLU function;

[0029] It should be noted that the node parameter of each node in the fully connected layer is 1*1*1, representing the two-dimensional space and one-dimensional spectrum respectively.

[0030] It is worth noting that the number of input nodes of the fully connected layer can also be 128, 256, 512, etc., which can be set according to the target to be detected and is not limited here.

[0031] In an embodiment of the present application, before step S20, the hyperspectral image target detection method based on deep learning may further include the following execution steps:

[0032] S41, using the atmospheric correction algorithm and the geometric correction algorithm to process the Pavia University dataset to obtain a preprocessed dataset;

[0033] S42, training multiple three-dimensional convolutional neural networks using the preprocessed data set, and obtaining different overall accuracy values ​​by changing parameters of the multiple three-dimensional convolutional neural networks, wherein the parameters of the three-dimensional convolutional neural network include the number of iterations, the convolution kernel size, and the random discard value;

[0034] For example, the preprocessed dataset in this application is the Pavia University hyperspectral image dataset, which was obtained by imaging the urban area of ​​Pavia, Italy in 2003 using the German airborne reflective optical spectral imager (Reflective Optics Spectrographic Imaging System, ROSIS-03 for short). The sensor continuously images the ground in the wavelength range of 0.43-0.86 microns, covering visible and near-infrared bands. The image size is 610×340 pixels, and each pixel contains information from 103 spectral bands, providing rich spectral and spatial details.

[0035] Figure 2 This is a false color image of the Pavia University dataset. The dataset has been atmospherically corrected but not geometrically corrected, with a spatial resolution of 1.3 meters. The original data contains 115 bands, but 12 of them were affected by noise and removed before use, so the remaining 103 bands were used. The Pavia University dataset is divided into 9 different ground object categories, including asphalt, grass, gravel, trees, metal plates, bare soil, asphalt, bricks, and shadows, etc., with a total of 42,776 labeled samples. Table 1-1 is a list of ground object categories in the Pavia University dataset.

[0036] Table 1-2 Pavia University dataset feature categories

[0037]

[0038] In this experiment, due to the small amount of initial experimental data, the initial experimental results lack stability and persuasiveness. To ensure the stability and persuasiveness of the experimental results, a large number of experiments were conducted on the designed 3D-CNN architecture before the formal experiment. The architecture of 3D-CNN can be referenced Figure 5 Using the overall accuracy (OA) mentioned above as the evaluation criterion, the parameters that have a greater impact on the experimental results, such as the number of iterations, convolution kernel size, Dropout value, etc., are adjusted and optimized to achieve the best experimental results.

[0039] In terms of the number of iterations, the training results are set to 100, 200, 300, 400, and 500 rounds, as shown in Table 4-2:

[0040] Table 1-3 OA values ​​of Pavia University dataset at different iteration times (%)

[0041]

[0042] It can be seen from Table 1-2 that as the number of iterations increases, OA approaches a stable value, but the model training time continues to increase. Therefore, considering the model training time and OA value, the number of iterations in this section is selected as 200.

[0043] In terms of convolution kernel size, we set convolution kernels of different sizes such as 2×2, 3×3, and 4×4 for training, and the convolution kernel spectral dimension is uniformly set to 7. The training results are shown in Table 1-3:

[0044] Table 1-4 OA values ​​of different convolution kernel sizes of Pavia University dataset (%)

[0045]

[0046] From Table 1-3, we can see that when the convolution kernel size is set to 3×3, OA can get the highest value, that is, the model has the highest recognition accuracy. Therefore, the convolution kernel size is set to 3×3 in this experiment.

[0047] In terms of adjusting the Dropout value, the Dropout value is set to 0.4, 0.5, 0.6, and 0.7 for training. The results are shown in Table 1-4:

[0048] Table 1-5 OA values ​​(%) for different Dropout values ​​of Pavia University dataset

[0049]

[0050] It can be seen from Table 1-4 that when the Dropout value is 0.6, the OA accuracy is the highest, which can effectively avoid overfitting.

[0051] According to the experimental results of parameter adjustment, this section sets the number of training iterations to 200, the convolution kernel size to 3×3, and the Dropout value to 0.6.

[0052] S43. Select the three-dimensional convolutional neural network corresponding to the maximum overall accuracy value as the three-dimensional convolutional neural network to be evaluated.

[0053] In an embodiment of the present application, after step S43, the hyperspectral image target detection method based on deep learning may further include the following execution steps:

[0054] S44. Evaluate the three-dimensional convolutional neural network to be evaluated based on the confusion matrix of the spectral remote sensing image to obtain a pre-constructed three-dimensional convolutional neural network.

[0055] For example, the confusion matrix of this experiment based on 3D-CNN using the Pavia University hyperspectral image dataset is shown in Table 1-5 below:

[0056] Table 1-6 Pavia University dataset confusion matrix

[0057]

[0058]

[0059] The names of various types of features are replaced by numbers, and the specific correspondence is shown in Table 4-6. As can be seen from Table 4-5, most samples are correctly identified, and only a small number of samples are incorrectly identified into a few categories. From the confusion matrix, the OA value is 95.070%, the AA value is 96.5%, and the Kappa coefficient is 0.936.

[0060] According to the calculation results, it can be concluded that 3D-CNN can effectively identify and classify ground objects based on hyperspectral datasets. At the same time, the AA and OA values ​​are both above 95%, and the model recognition accuracy is high.

[0061] The recognition accuracy of various types of objects is shown in Table 1-6:

[0062] Table 1-7 Pavia University dataset object recognition accuracy

[0063]

[0064] As can be seen from Table 1-6, excluding the undefined areas, the 3D-CNN model training results for all types of objects have reached more than 90% recognition results, among which the recognition effect for metal plates is the best, with a recognition accuracy of 99.9%. Secondly, the recognition accuracy for unobstructed areas and shadows is also more than 98%. The model is very accurate in identifying various types of objects and can effectively distinguish between various types of objects. The experimental results are shown in the figure below. Figure 3 As shown, from Figure 2 and Figure 3 It can be seen intuitively from the comparison that the 3D-CNN model training results have high accuracy and can accurately identify multiple types of objects with similar spatial features, such as asphalt and asphalt.

[0065] S30: Output the target image after classification and detection.

[0066] Based on the above embodiments, the present application also provides a hyperspectral image target detection device based on deep learning, referring to Figure 4 The hyperspectral image target detection device 100 based on deep learning may include an acquisition module 101, a processing module 102, and an output module 103, wherein the acquisition module 101 can be used to acquire the preprocessed spectral remote sensing image; the processing module 102 can be used to process the preprocessed spectral remote sensing image based on a pre-built three-dimensional convolutional neural network to obtain a target image after classification detection; wherein the three-dimensional convolutional neural network includes a first three-dimensional convolutional layer, a first activation layer, a second three-dimensional convolutional layer, a second activation layer and a fully connected layer connected in sequence, and the convolution kernel sizes of the first three-dimensional convolutional layer and the second three-dimensional convolutional layer are both 337, that is, the spatial dimension of the convolution kernel is 3, the spectral dimension of the convolution kernel is 7, the number of input nodes of the fully connected layer is 64, the number of output nodes is 1, and the first activation layer and the second activation layer both use ReLU function; the output module 103 can be used to output the target image after classification detection.

[0067] It is not difficult to find that this embodiment is a system embodiment corresponding to the above method embodiment, and this embodiment can be implemented in conjunction with the above method embodiment. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiments.

[0068] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.

[0069] Based on the above embodiments, the present application also provides an electronic device, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the hyperspectral image target detection method based on deep learning provided in any of the above embodiments.

[0070] Among them, the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripheral devices, voltage regulators and power management circuits together, which are all well known in the art, so this article will not describe them further. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on the wireless medium through the antenna, and further, the antenna also receives the data and transmits the data to the processor.

[0071] The processor is responsible for managing the bus and general processing, and may also provide various functions, including timing, peripheral interfacing, voltage regulation, power management, and other control functions. Magnetic storage may be used to store data used by the processor when performing operations.

[0072] Based on the above embodiments, the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement the hyperspectral image target detection method based on deep learning provided by any of the above embodiments.

[0073] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM (Read-Only Memory), RAM (Random Access Memory), disk or optical disk and other media that can store program codes.

[0074] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A hyperspectral image target detection method based on deep learning, characterized in that: include: Acquire preprocessed spectral remote sensing images; Processing the preprocessed spectral remote sensing image based on a prebuilt three-dimensional convolutional neural network to obtain a target image after classification detection, wherein the three-dimensional convolutional neural network includes a first three-dimensional convolutional layer, a first activation layer, a second three-dimensional convolutional layer, a second activation layer and a fully connected layer connected in sequence, the convolution kernel sizes of the first three-dimensional convolutional layer and the second three-dimensional convolutional layer are both 3×3×7, that is, the spatial dimension of the convolution kernel is 3, the spectral dimension of the convolution kernel is 7, the number of input nodes of the fully connected layer is 64, the number of output nodes is 1, and the first activation layer and the second activation layer both use the ReLU function; Output the target image after classification detection.

2. The method for hyperspectral image target detection based on deep learning according to claim 1, characterized in that: Before the spectral remote sensing image is processed based on the pre-built three-dimensional convolutional neural network to obtain the target image after classification detection, the hyperspectral image target detection method based on deep learning also includes: The Pavia University dataset was processed using the atmospheric correction algorithm and the geometric correction algorithm to obtain the preprocessed dataset; Training a plurality of the three-dimensional convolutional neural networks using the preprocessed data set, and obtaining different overall accuracy values ​​by changing parameters of the plurality of the three-dimensional convolutional neural networks, wherein the parameters of the three-dimensional convolutional neural network include the number of iterations, the convolution kernel size, and the random discard value; The three-dimensional convolutional neural network corresponding to the maximum overall accuracy value is selected as the three-dimensional convolutional neural network to be evaluated.

3. The method for hyperspectral image target detection based on deep learning according to claim 2 is characterized in that: After selecting the three-dimensional convolutional neural network corresponding to the maximum overall accuracy value as the three-dimensional convolutional neural network to be evaluated, the hyperspectral image target detection method based on deep learning further includes: The three-dimensional convolutional neural network to be evaluated is evaluated based on the confusion matrix of the spectral remote sensing image to obtain the pre-constructed three-dimensional convolutional neural network.

4. The method for hyperspectral image target detection based on deep learning according to claim 1, characterized in that: The step of obtaining the preprocessed spectral remote sensing image comprises: Acquire spectral remote sensing images; The spectral remote sensing image is processed based on a radiation calibration algorithm, a histogram equalization algorithm and a median filter to obtain the preprocessed spectral remote sensing image.

5. A hyperspectral image target detection device based on deep learning, characterized in that: include: An acquisition module is used to acquire the preprocessed spectral remote sensing image; A processing module, used for processing the preprocessed spectral remote sensing image based on a prebuilt three-dimensional convolutional neural network to obtain a target image after classification detection, wherein the three-dimensional convolutional neural network includes a first three-dimensional convolutional layer, a first activation layer, a second three-dimensional convolutional layer, a second activation layer and a fully connected layer connected in sequence, the convolution kernel sizes of the first three-dimensional convolutional layer and the second three-dimensional convolutional layer are both 337, that is, the spatial dimension of the convolution kernel is 3, the spectral dimension of the convolution kernel is 7, the number of input nodes of the fully connected layer is 64, the number of output nodes is 1, and the first activation layer and the second activation layer both use ReLU function; The output module is used to output the target image after classification detection.

6. An electronic device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the hyperspectral image target detection method based on deep learning as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for hyperspectral image target detection based on deep learning as described in any one of claims 1 to 4 can be implemented.

Images