Remote sensing image target detection method based on wavelet convolution sequence network

Abstract

The invention provides a remote sensing image target detection method based on a wavelet convolution sequence network, and the method comprises the steps: applying wavelet transformation to a backbone network of a target detection model of a remote sensing image, and extracting attribute features of different dimensions for the same target through the extraction capability of the wavelet transformation for different frequency features; a wavelet convolution sequence module and a convolution sequence fusion module are used to reduce the amount of calculation.

Description

Technical Field

[0001] The present invention belongs to the technical fields of information processing, remote sensing image target detection, etc., and specifically relates to a remote sensing image target detection method based on a wavelet convolution sequence network, which can be applied to the fields of agricultural water resources and forest resources management, urban planning management, and environmental and ocean detection. Background Art

[0002] As a key technology in remote sensing image applications, remote sensing image target detection is centered on accurately identifying and locating specific objects in images. This is not only crucial to improving the application value of remote sensing data, but also through accurate target detection, it can effectively support decision-making in areas such as resource management, environmental monitoring, and disaster response, thereby providing important support for sustainable development and environmental and ecological security.

[0003] The Chinese patent "A remote sensing image target detection method based on a fused convolutional attention mechanism" (CN116229295A) aims to solve the problems of long training time caused by the difficulty of convergence of the attention mechanism in the end-to-end target detection framework of remote sensing images based on DETR-based Transformer, and the low accuracy of small target detection caused by the inability of the attention mechanism to effectively obtain local information. A remote sensing image target detection method based on a fused convolutional attention mechanism with a stronger ability to obtain local features of the image and faster model convergence speed is proposed. However, it still belongs to the target detection model based on Transformer. The number of parameters and the amount of calculation of the model make it unable to meet the target detection needs of remote sensing images in scenarios with high real-time requirements. Xie et al. in the literature "Xingxing Xie, Gong Cheng, Jiabao Wang, XiwenYao, Junwei Han, "Oriented R-CNN for Object Detection,"

[0004] International Conference on Computer Vision (ICCV), 2021, pp. 3520-3529." The regression method of the rotated bounding box was updated, and the original regression method of 4 points and 8 coordinate values ​​was changed to a method of 6 values ​​in total, which is the 4 values ​​of the bounding box plus the 2 values ​​of the midpoint coordinate offset. This reduces the computational cost of the bounding box representation, frees up the computing resources that originally relied on placing a large number of rotated anchor boxes, and improves the computational efficiency of the regressed bounding box. However, it still does not solve the problem that small target features are difficult to extract in remote sensing images, and it is difficult to meet the target detection requirements of remote sensing images in scenarios with high accuracy requirements. Feng et al. in the document "Mingkui Feng, Hancheng Yu, Xiaoyu Dang and Ming Zhou. "Category-Aware Dynamic Label Assignment with High-Quality Oriented Proposal." ArXiv abs / 2407.03205(2024):n.pag." In this paper, the angle representation of the rotating bounding box is introduced into the complex plane, which reduces the possibility of discontinuity of the label regression value at the boundary caused by the periodicity of the angle and avoids sudden fluctuations caused by the loss function. However, this method does not solve the problem of decreased target detection accuracy caused by interference information such as occlusion and blur, and it is difficult to meet the target detection needs of remote sensing images in scenarios with high anti-interference requirements. Summary of the invention

[0005] The purpose of the present invention is to propose a remote sensing image target detection method based on a wavelet convolution sequence network in response to the defects and shortcomings of the above-mentioned existing solutions. The method applies wavelet transform to the backbone network of the target detector using remote sensing images, mainly solving the problems of low detection accuracy and slow detection rate caused by interference such as occlusion and noise in complex environments. The detection implementation steps include: (1) constructing a training data set, a verification data set and a test data set; (2) constructing a target detection model based on a wavelet convolution sequence network; (3) training the model based on a large number of training data sets; (4) detecting the target based on the test data set and outputting the final prediction result. The network model constructed by the present invention uses a wavelet convolution module and a convolution sequence fusion module to reduce the amount of calculation, significantly improve the target detection speed, and have high detection accuracy.

[0006] The technical solution specifically adopted by the present invention to solve the technical problem is:

[0007] A remote sensing image target detection method based on wavelet convolution sequence network applies wavelet transform to the backbone network of the remote sensing image target detection model. Through the ability of wavelet transform to extract different frequency features, attribute features of different dimensions are extracted for the same target. Wavelet convolution sequence module and convolution sequence fusion module are used to reduce the amount of calculation.

[0008] Furthermore, the process of constructing a training data set, a validation data set and a test data set for the target detection model includes:

[0009] Scaling all images in the database using a bilinear interpolation algorithm with a size ratio of [R1, R2, R3] to obtain image blocks of different sizes;

[0010] The optical remote sensing images containing the labeled targets in the database are cut into blocks with a rectangular window of size Hc×Hc and an overlapping gap of size O, and the data blocks after cutting are obtained with a size of Hc×Hc×3, and the overlapping pixels between the data blocks are O;

[0011] All the cut data blocks are divided into a training data set T consisting of N1 data blocks, a verification data set V consisting of N2 data blocks, and a test data set U consisting of N3 data blocks according to the size ratio of [R4, R5, R6].

[0012] Furthermore, in the target detection model, a wavelet convolution sequence network is constructed by fusing a wavelet convolution sequence module with a convolution sequence fusion module without increasing the number of parameters and expanding the receptive field.

[0013] Furthermore, the wavelet convolution sequence module includes two wavelet convolution operations WTconv; the wavelet convolution operation WTconv includes at least two branches, wherein the first branch only performs a convolution operation on the image, and the second branch obtains four images through discrete wavelet transform, and after the convolution operation and the inverse wavelet transform operation, the images are element-wise added with the image of the first branch; if the number of branches exceeds 2, the processing process of the nth branch (n>2) is the same as that of the second branch, and the n-1th branch is element-wise added before the inverse wavelet transform operation of the n-1th branch, and the outputs of all branches are output after the element-wise addition operation is completed.

[0014] Furthermore, the discrete wavelet transform is a Haar wavelet transform.

[0015] Furthermore, in the wavelet convolution sequence network, for a given image X, it is input into a wavelet convolution operation WTconv_1 to obtain a feature map F1, and then the feature map F1 is input into another wavelet convolution operation WTconv_2 to obtain a feature map F2; F1 and F2 are input into the convolution sequence fusion module to obtain a feature map S; and then the feature map S is added to the input image X to obtain the final feature map Y.

[0016] Furthermore, the specific process of the convolution sequence fusion module is as follows: the feature map F1 and the feature map F2 are respectively processed by a 1×1 convolution Get feature maps of the same dimension Will After passing through an average pooling layer, a maximum pooling layer, and then inputting into a weight distribution layer, the feature map SA is obtained; the feature map SA is respectively connected to the feature map and Multiply them, and add the multiplication results to get the feature map S.

[0017] Furthermore, in the target detection model, the backbone network extracts target features of remote sensing images by a wavelet convolution sequence network, the neck network integrates feature images of different dimensions extracted by the backbone network by an FPN network, and the head network obtains the target area candidate box and the final classification score by an OrientedRCNN.

[0018] Furthermore, the detection process is specifically as follows: the data blocks in the test data set are input into the trained target detection model based on the wavelet convolution sequence network in turn to obtain the candidate box of each data block in the test data set, the score corresponding to the candidate box, and the target category corresponding to the candidate box; all candidate boxes of the target category with a score higher than 0.5 are retained, and the remaining candidate boxes are discarded; finally, non-maximum suppression is performed on all the target bounding box coordinates and confidences obtained after retention to obtain the detection results of the remote sensing image in the test data set.

[0019] And, an electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the remote sensing image target detection method based on the wavelet convolution sequence network as described above are implemented.

[0020] A non-transitory computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the remote sensing image target detection method based on a wavelet convolution sequence network as described above.

[0021] Compared with the prior art, the present invention and its preferred solution introduce wavelet transform into the feature extraction network of target detection. Through the ability of wavelet transform to extract features of different frequencies, attribute features of different dimensions are extracted for the same target, and the size of the receptive field in target detection is guaranteed without increasing the computing resources used, thereby achieving better results. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0023] Figure 1 It is an overall flow chart of the embodiment of the present invention;

[0024] Figure 2 It is a wavelet convolution module and a wavelet convolution sequence network diagram in an embodiment of the present invention;

[0025] Figure 3 This is a framework diagram of a target detection model based on a wavelet convolutional sequence network in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the features and advantages of the present invention more clearly understood, the following embodiments are specifically described in detail as follows:

[0027] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0029] like Figure 1 As shown, the steps implemented by the present invention are as follows:

[0030] Step 1: Build training data set, validation data set and test data set:

[0031] a. Perform a scaling operation on all images in the database using a bilinear interpolation algorithm with a size of [R1, R2, R3] to obtain image blocks of different scales. In the embodiment of the present invention, the database is provided by the DOTA optical remote sensing image dataset, R1 = 0.5, R2 = 1.0, R3 = 1.5, and the target bounding box coordinates are composed of the upper left corner coordinates and the lower right corner coordinates of the target;

[0032] b. Using a rectangular window of size Hc×Hc and an overlapping gap of size O, the optical remote sensing image containing the marked target in the database is cut into blocks, and the data blocks after cutting are obtained with a size of Hc×Hc×3, and the overlapping pixels between the data blocks are O. In the embodiment of the present invention, Hc=1024, O=500;

[0033] c. Name each cut image block according to the dataset naming rules. The dataset naming rules refer to connecting the file name of each remote sensing image to be cut, the number of sliding window steps corresponding to the cut data block, the initial position of the sliding window x-axis, and the initial position of the sliding window y-axis with the English dash “--” symbol to generate a .jpg format file.

[0034] d. Divide all the cut data blocks into a training data set T consisting of N1 data blocks, a verification data set V consisting of N2 data blocks, and a test data set U consisting of N3 data blocks. In this embodiment of the present invention, R4 = 0.5, R5 = 0.17, R6 = 0.33, N1 = 138883, N2 = 35462, and N3 = 71888.

[0035] Step 2: Build a target detection model based on wavelet convolution sequence network:

[0036] a. Construct a wavelet convolution module for extracting image features;

[0037] In the wavelet convolution sequence module of this embodiment, the Haar wavelet transform operation is preferably used because it is computationally efficient and simple. However, it should be noted that the implementation of the solution of the present invention is not limited to the Haar wavelet, and other types of discrete wavelet bases can also be used, although the computational cost will increase.

[0038] like Figure 2 As shown in the figure, first, the image X needs to go through two branches. One branch only performs convolution operation on the image X. The second branch obtains four images X through Haar wavelet. LL (1) , X LH (1) , X HL (1) , X HH (1) , the four images need to undergo convolution and inverse wavelet transform operations before being added to the image of the first branch. If the second branch needs to perform two Haar wavelets, then X LL (2) …and so on. Then the four images are also subjected to convolution operation and inverse wavelet transform operation, and the images after convolution operation in the two branches are added together.

[0039] In general, a third branch or even a fourth branch can be extended after the second branch, which is inferred according to the number of Haar wavelets performed.

[0040] Taking Haar wavelet as an example, the process of performing a Haar wavelet (WT) includes: given an image X, a first-level HaarWT on a spatial dimension (width or height) is composed of kernel and is given by a depthwise convolution of , followed by a standard downsampling operator by a factor of 2. To perform 2D Haar WT, the operations are combined in two dimensions, using the following four filters, resulting in a depthwise convolution with a stride of 2:

[0041]

[0042] Among them, f LL is a low-pass filter, and f LH 、f HL 、f HH is a set of high-pass filters. For each input channel, the output of the convolution satisfies the following formula:

[0043] [X LL ,X LH ,X HL ,X HH ]=Conv([f LL ,f LH ,f HL ,f HH ],X) (2)

[0044] The left side of the above formula contains four channels, and the resolution of each channel (in each spatial dimension) is half of X. LL represents the low-frequency component of X, while X LH , X HL , X HH Represent its horizontal, vertical and diagonal high-frequency components respectively.

[0045] Since the kernel in the above formula forms a standard orthogonal basis, the inverse wavelet transform (IWT) is obtained by transposed convolution:

[0046] X=Conv-transposed([f LL ,f LH ,f HL ,f HH ],[X LL ,X LH ,X HL ,X HH ]) (3)

[0047] Then the cascaded wavelet decomposition is given by recursively decomposing the low-frequency components, and each layer of the decomposition is:

[0048]

[0049] in i is the index of the current level. This process results in an increase in frequency resolution and a decrease in low-frequency spatial resolution.

[0050] Consider that increasing the kernel size of the spatial convolutional layer will increase the number of parameters (number of degrees of freedom) in a quadratic manner. To mitigate this. In this embodiment, the low-frequency and high-frequency content of the input is first filtered and reduced using a wavelet transform. Then, a small kernel depth convolution is performed on the different frequency maps before constructing the output using the IWT. The process is as follows:

[0051] Y=IWT(Conv(W,WT(X))) (5)

[0052] Where X is the input tensor and W is the weight tensor of the k×k deep kernel, whose number of input channels is four times that of X. This operation not only separates the convolutions between frequency components, but also allows smaller kernels to operate in a larger area of ​​the original input, i.e., increases its receptive field input.

[0053] like Figure 2 As shown. The derivation process of performing multiple Haar wavelets in the wavelet convolution operation and the process of calculating each component are as follows: The wavelet convolution module of the embodiment of the present invention adopts a one-level combination operation of WT and convolution, and further increases the number of Haar wavelets by using the same cascade principle as formula (4). The process is given by the following formula:

[0054]

[0055] in is the input of this layer, represents all three high-frequency maps of the i-th layer described above. In order to combine the outputs of different frequencies, this embodiment uses the property that WT and its inverse are linear operations, that is,

[0056] IWT(X+Y)=IWT(X)+IWT(Y). Therefore, the following formula is executed:

[0057]

[0058] To get the sum of different levels of convolution, where Z (i) is the summary output starting from layer i. The output is the sum of two convolution outputs of different sizes.

[0059] In theory, we would like to perform normalization after convolution on each frequency component, however, since we are working in the wavelet domain, the individual normalization of each frequency map does not correspond to the normalization in the original domain. Therefore, we found that instead of normalization, only performing channel scaling to measure the contribution of each frequency component can achieve the desired effect.

[0060] b. Without increasing the number of parameters and expanding the receptive field, the wavelet convolution sequence module and the convolution sequence fusion module are integrated to construct a wavelet convolution sequence network;

[0061] like Figure 2 As shown, in the wavelet convolution sequence network of this embodiment, for a given image X, it is input into a wavelet convolution operation WTconv_1 to obtain a feature map F1, and then the feature map F1 is input into another wavelet convolution operation WTconv_2 to obtain a feature map F2. F1 and F2 are input into the convolution sequence fusion module to obtain a feature map S. The feature map S is added to the input image X to obtain the final feature map Y.

[0062] The specific steps in the convolution sequence fusion module are as follows: F1 and F2 are respectively combined with a 1×1 convolution Get the same dimension feature map Will After being connected to an average pooling layer and then a maximum pooling layer, and then input into a weight distribution layer, the feature map SA is obtained. The feature map SA is respectively connected to the feature map and Multiply them, and add the multiplication results to get the feature map S.

[0063] In summary, the wavelet convolution sequence module composed of the connected WTconv_1 and WTconv_2 and the convolution sequence fusion module constitute a wavelet convolution sequence network.

[0064] c. Build a complete target detection model;

[0065] like Figure 3As shown, in the target detection model of the present invention, the backbone network extracts the target features of the remote sensing image by a wavelet convolution sequence network, the neck network integrates the different dimensional feature images extracted by the backbone network by the FPN network, and finally the head network obtains the target area candidate box and the final classification score by the OrientedRCNN, that is, the candidate box coordinates, the score corresponding to the candidate box and the category corresponding to the candidate box are obtained by the regression classification layer. Among them, the FPN network and the OrientedRCNN network are both prior arts. The general idea of ​​the FPN method is: to integrate the feature maps of different scales that continuously pass through the convolution layer in the backbone network, and in this way to obtain target feature information of different scales. The general idea of ​​the OrientedRCNN method is: by increasing the parameters required for prediction of the regression classification layer, the predicted bounding box is converted into a method of predicting a rotated bounding box within the bounding box, thereby reducing the problem of low detection rate of the remote sensing image target detection model caused by the preset anchor point.

[0066] Step 3: Train the model based on a large number of training data sets;

[0067] Step 4: Detect the target according to the test data set and output the final prediction result:

[0068] a. Input the data blocks in the test data set into the trained target detection model based on wavelet convolution sequence network in sequence, and obtain the candidate box, the score corresponding to the candidate box, and the target category corresponding to the candidate box for each data block in the test data set;

[0069] b. Keep all candidate boxes of target categories with scores higher than 0.5 and discard the rest;

[0070] c. Perform non-maximum suppression on all target bounding box coordinates and confidences obtained after retention to obtain the detection results of remote sensing images in the test data set. Among them, non-maximum suppression is an existing technology. The general idea of ​​this method is: sort all bounding box coordinates from high to low according to confidence, retain bounding boxes with low overlap with other bounding boxes and high confidence, and discard bounding boxes with high overlap with other bounding boxes and low confidence.

[0071] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0072] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0075] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0076] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

[0077] The present invention is not limited to the above-mentioned optimal implementation mode. Anyone can derive other various forms of remote sensing image target detection methods based on wavelet convolution sequence networks under the inspiration of the present invention. All equal changes and modifications made within the scope of the invention application of the present invention should fall within the scope of the present invention.

Claims

1. A remote sensing image target detection method based on wavelet convolution sequence network, characterized in that: The wavelet transform is applied to the backbone network of the target detection model of remote sensing images. Through the ability of wavelet transform to extract features of different frequencies, attribute features of different dimensions are extracted for the same target. The wavelet convolution sequence module and the convolution sequence fusion module are used to reduce the amount of calculation.

2. The remote sensing image target detection method based on wavelet convolution sequence network according to claim 1 is characterized in that: The process of constructing a training data set, a validation data set and a test data set for the target detection model includes: Scaling all images in the database using a bilinear interpolation algorithm with a size ratio of [R1, R2, R3] to obtain image blocks of different sizes; The optical remote sensing images containing the labeled targets in the database are cut into blocks with a rectangular window of size Hc×Hc and an overlapping gap of size O, and the data blocks after cutting are obtained with a size of Hc×Hc×3, and the overlapping pixels between the data blocks are O; All the cut data blocks are divided into a training data set T consisting of N1 data blocks, a verification data set V consisting of N2 data blocks, and a test data set U consisting of N3 data blocks according to the size ratio of [R4, R5, R6].

3. The remote sensing image target detection method based on wavelet convolution sequence network according to claim 1 is characterized in that: In the target detection model, a wavelet convolution sequence network is constructed by fusing a wavelet convolution sequence module with a convolution sequence fusion module without increasing the number of parameters and expanding the receptive field.

4. The remote sensing image target detection method based on wavelet convolution sequence network according to claim 1 is characterized in that: The wavelet convolution sequence module includes two wavelet convolution operations WTconv; the wavelet convolution operation WTconv includes at least two branches, wherein the first branch only performs a convolution operation on the image, and the second branch obtains four images through discrete wavelet transform, and after the convolution operation and the inverse wavelet transform operation, the images are element-wise added with the images of the first branch; if the number of branches exceeds 2, the processing process of the nth branch (n>2) is the same as that of the second branch, and the elements are added with the n-1th branch before the inverse wavelet transform operation of the n-1th branch, and the outputs of all branches are output after the element-wise addition operation is completed.

5. The remote sensing image target detection method based on wavelet convolution sequence network according to claim 4 is characterized in that: The discrete wavelet transform is Haar wavelet transform.

6. The remote sensing image target detection method based on wavelet convolution sequence network according to claim 3 is characterized in that: In the wavelet convolution sequence network, for a given image X, it is input into a wavelet convolution operation WTconv_1 to obtain a feature map F1, and then the feature map F1 is input into another wavelet convolution operation WTconv_2 to obtain a feature map F2; F1 and F2 are input into the convolution sequence fusion module to obtain a feature map S; and then the feature map S is added to the input image X to obtain the final feature map Y.

7. The remote sensing image target detection method based on wavelet convolution sequence network according to claim 6 is characterized in that: The specific process of the convolution sequence fusion module is as follows: the feature map F1 and the feature map F2 are respectively processed by a 1×1 convolution F1 1×1 、F2 1×1 Get feature maps of the same dimension Will After being connected, passing through an average pooling layer, a maximum pooling layer, and then inputting into a weight distribution layer, a feature map SA is obtained; The feature map SA is respectively and Multiply them, and add the multiplication results to get the feature map S.

8. The remote sensing image target detection method based on wavelet convolution sequence network according to claim 3 is characterized by: In the target detection model, the backbone network extracts target features of remote sensing images by wavelet convolution sequence network, the neck network integrates different dimensional feature images extracted by the backbone network by FPN network, and the head network obtains the target area candidate box and the final classification score by OrientedRCNN.

9. The remote sensing image target detection method based on wavelet convolution sequence network according to claim 8 is characterized in that: The specific detection process is as follows: the data blocks in the test data set are sequentially input into the trained target detection model based on the wavelet convolution sequence network to obtain the candidate box, the score corresponding to the candidate box, and the target category corresponding to the candidate box for each data block in the test data set; all candidate boxes of the target category with a score higher than 0.5 are retained, and the rest of the candidate boxes are discarded; Finally, all the retained target bounding box coordinates and confidence scores are subjected to non-maximum suppression to obtain the detection results of the remote sensing images in the test dataset.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the remote sensing image target detection method based on wavelet convolution sequence network as described in any one of claims 1 to 9 are implemented.

Images