A remote sensing image target detection method and system based on frequency domain feature reconstruction

By introducing a frequency domain feature reconstruction module into the remote sensing image object detection model, the feature extraction and fusion are optimized, and the problem of low accuracy of target detection of scale changes is solved, which significantly improves the multi-dimensional information perception ability and detection accuracy of remote sensing image object detection.

CN118154843BActive Publication Date: 2024-11-22耕宇牧星(北京)空间科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410220317.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-02-28
Publication Date
2024-11-22
Estimated Expiration
2044-02-28

AI Technical Summary

Technical Problem

The existing remote sensing image object detection model performs poorly when processing scale changes targets and lacks feature extraction capabilities, resulting in low detection accuracy.

Method used

Using the method based on frequency domain feature reconstruction, the frequency domain feature reconstruction module is constructed, including Fourier transform, inverse Fourier transform, multi-layer perceptron, upsampling, self-attention and point convolution modules, for feature fusion network In this paper, feature extraction and fusion are optimized to capture multi-scale feature information.

Benefits of technology

The model's perception ability of multi-dimensional information is improved, and multi-scale feature information of potential target areas can be captured more accurately, realizing accurate remote sensing image object detection task recognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118154843B_ABST
    Figure CN118154843B_ABST
Patent Text Reader

Abstract

The present invention discloses a remote sensing image target detection method and system based on frequency domain feature reconstruction, belonging to the field of image processing technology. The method comprises: constructing a frequency domain feature reconstruction module; and building a feature fusion network optimized based on the frequency domain feature reconstruction module and its corresponding remote sensing image target detection model; then training the remote sensing image target detection model for target detection tasks; and finally using the trained remote sensing image target detection model to perform target detection on the remote sensing image to be tested. The present invention can improve the model's ability to perceive multi-dimensional information of typical remote sensing image scale-changing targets during the feature fusion process, and facilitate the model to better capture multi-scale feature information about potential target areas from a frequency domain perspective, which is conducive to the realization of accurate remote sensing image target detection task recognition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a remote sensing image target detection method and system based on frequency domain feature reconstruction. Background Art

[0002] Remote sensing image interpretation has a wide range of applications in the fields of national defense security, environmental monitoring, and urban planning, and remote sensing image target detection technology has always been one of the key research directions. At present, the mainstream remote sensing image target detection methods are mostly based on the construction of convolutional neural networks, which output the prediction box and type information of the potential target area through steps such as feature extraction, feature fusion, and feature decoupling. In particular, the feature fusion network occupies an important position in the entire target detection model and is one of the core structures of the model. However, the current mainstream target detection model mainly adopts a construction method based on natural images, and the scale of targets in remote sensing images varies dramatically. For example, aircraft and ships have large scale differences and difficult feature extraction. This has led to the poor performance of traditional target detection models based on convolutional neural networks in remote sensing image interpretation.

[0003] Therefore, it is urgent to overcome the inadequate ability of feature extraction networks in dealing with targets of varying scales and to improve remote sensing image target detection methods in order to achieve accurate remote sensing image target detection task recognition. Summary of the invention

[0004] In view of this, the present invention provides a remote sensing image target detection method and system based on frequency domain feature reconstruction that can solve the above-mentioned technical problems, which is conducive to achieving accurate remote sensing image target detection task recognition.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] In a first aspect, an embodiment of the present invention provides a remote sensing image target detection method based on frequency domain feature reconstruction, the method comprising the following steps:

[0007] S1, construct frequency domain feature reconstruction module;

[0008] S2. Building a feature fusion network and a remote sensing image target detection model based on the frequency domain feature reconstruction module;

[0009] S3, performing target detection task training on the remote sensing image target detection model;

[0010] S4. Use the trained remote sensing image target detection model to perform target detection on the remote sensing image to be tested.

[0011] Furthermore, in step S1, the frequency domain feature reconstruction module constructed includes: a Fourier transform module, an inverse Fourier transform module, a feature splicing module, a multi-layer perceptron module, an upsampling module, a self-attention module and a point convolution module, wherein:

[0012] The Fourier transform module is used to map the feature map to the frequency domain space to extract the amplitude vector and the phase vector; the inverse Fourier transform module is used to restore the frequency domain space feature map back to the original time domain space; the multi-layer perceptron module is used to learn the effective feature information of the frequency domain vector and adjust its dimension to an appropriate size; the upsampling module and the feature splicing module are used to adjust the shape of the feature map; the self-attention module is used to establish long-distance feature dependencies to capture global information and optimize the information quality of the feature map.

[0013] Furthermore, the up-sampling module uses a bilinear interpolation method to adjust the amplitude vector to the size of the phase vector.

[0014] Furthermore, in step S2, the remote sensing image target detection model constructed includes: a backbone feature extraction network, a feature fusion network optimized based on a frequency domain feature reconstruction module, and a feature decoupling head.

[0015] Furthermore, the backbone feature extraction network is composed of a plurality of convolution modules stacked together, each of which includes: a basic convolution operator, a normalization module and an activation function, wherein:

[0016] The basic convolution operator calculates the channel information of the input remote sensing image at a pixel-by-pixel level; the normalization module normalizes the calculated channel results to the standard range; the activation function activates and inhibits the feature matrix of the potential target area.

[0017] Furthermore, in step S3, focal loss, rotated frame intersection-over-union loss and cross entropy loss are used to train the remote sensing image target detection model for target detection tasks.

[0018] In a second aspect, an embodiment of the present invention further provides a remote sensing image target detection system based on frequency domain feature reconstruction, which applies the above-mentioned remote sensing image target detection method based on frequency domain feature reconstruction to achieve accurate remote sensing image target detection task recognition, and the system includes:

[0019] A target detection model building module is used to construct a frequency domain feature reconstruction module, and to build a feature fusion network and a remote sensing image target detection model based on the frequency domain feature reconstruction module;

[0020] A target detection model training module is used to perform target detection task training on the remote sensing image target detection model;

[0021] The target detection model application module is used to use the trained remote sensing image target detection model to perform target detection on the remote sensing image to be tested.

[0022] Furthermore, the constructed frequency domain feature reconstruction module includes: a Fourier transform module, an inverse Fourier transform module, a feature splicing module, a multi-layer perceptron module, an upsampling module, a self-attention module and a point convolution module, wherein:

[0023] The Fourier transform module is used to map the feature map to the frequency domain space to extract the amplitude vector and the phase vector; the inverse Fourier transform module is used to restore the frequency domain space feature map back to the original time domain space; the multi-layer perceptron module is used to learn the effective feature information of the frequency domain vector and adjust its dimension to an appropriate size; the upsampling module and the feature splicing module are used to adjust the shape of the feature map; the self-attention module is used to establish long-distance feature dependencies to capture global information and optimize the information quality of the feature map.

[0024] Compared with the prior art, the present invention has at least the following beneficial effects:

[0025] 1. The present invention provides a remote sensing image target detection method and system based on frequency domain feature reconstruction, wherein a frequency domain feature reconstruction module is built, and a feature fusion network optimized based on the frequency domain feature reconstruction module and its corresponding remote sensing image target detection model are constructed, which can improve the model's multi-dimensional information perception ability for typical remote sensing image scale-changing targets during the feature fusion process, and help the model better capture multi-scale feature information about potential target areas from a frequency domain perspective, which is conducive to the realization of accurate remote sensing image target detection task identification.

[0026] 2. The remote sensing image target detection model based on frequency domain feature reconstruction proposed in the present invention, in the feature fusion stage, uses the frequency domain feature component representation of feature maps at different scale levels to reconstruct additional feature information to assist prediction. The reconstructed feature maps supplement the dimension of the channel information under the original fusion path, helping the model to better capture the multi-scale feature information of the potential target area from a multi-dimensional perspective, which helps to achieve accurate remote sensing image target detection task recognition.

[0027] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0028] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0030] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0031] Figure 1 A flowchart of a remote sensing image target detection method based on frequency domain feature reconstruction is provided in an embodiment of the present invention.

[0032] Figure 2 A schematic diagram of the structure of a frequency domain feature reconstruction module provided in an embodiment of the present invention.

[0033] Figure 3 A schematic diagram of the structure of a remote sensing image target detection model based on frequency domain feature reconstruction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0035] In the description of the present invention, it should be noted that: in some processes described in the specification and drawings of this application, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may not be performed in the order in which they appear in this document or may be performed in parallel. In addition, various serial numbers are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0036] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] See also Figure 1 As shown, the present invention provides a remote sensing image target detection method based on frequency domain feature reconstruction, the method mainly comprises the following steps:

[0038] S1, construct frequency domain feature reconstruction module;

[0039] S2. Building a feature fusion network and a remote sensing image target detection model based on the frequency domain feature reconstruction module;

[0040] S3, performing target detection task training on the remote sensing image target detection model;

[0041] S4. Use the trained remote sensing image target detection model to perform target detection on the remote sensing image to be tested.

[0042] The specific implementation methods of the method of the present invention are described in detail below:

[0043] In this embodiment, an efficient frequency domain feature reconstruction module is designed to reconstruct additional feature information to assist prediction using the frequency domain feature component representation of feature maps at different scale levels. The reconstructed feature map supplements the dimension of the channel information under the original fusion path, helping the model to better capture the multi-scale feature information of the potential target area from a multi-dimensional perspective, and achieve accurate remote sensing image target detection task recognition. The specific implementation is as follows:

[0044] 1. Build the frequency domain feature reconstruction module

[0045] In this embodiment, a frequency domain feature reconstruction module is first constructed, and its structure is as follows: Figure 2 As shown in the figure, the frequency domain feature reconstruction module mainly includes sub-modules such as Fourier transform module, inverse Fourier transform module, feature splicing module, multi-layer perceptron module, upsampling module, self-attention module and point convolution module. Among them, the Fourier transform module is used to map the feature map to the frequency domain space to extract the amplitude vector and phase vector. The inverse Fourier transform module is used to restore the frequency domain space feature map back to the original time domain space. The multi-layer perceptron module is used to learn the effective feature information of the frequency domain vector and adjust its dimension to an appropriate size. The upsampling module and the feature splicing module are used to adjust the shape of the feature map, and the self-attention module is used to establish long-distance feature dependencies to capture global information and optimize the information quality of the feature map.

[0046] The calculation process will be described in detail below according to the module calculation order:

[0047] First, define the input feature map as ,in Indicates channel information. Represents the shape of the feature map. and The feature maps of adjacent layers of the frequency domain feature reconstruction module are input into the backbone feature extraction network, and the phase vector and amplitude vector are obtained by the fast Fourier transform module respectively. The calculation process can be formally expressed as:

[0048]

[0049] in, represents the fast Fourier transform module, Representation feature map The phase vector and amplitude vector are obtained through the fast Fourier transform module. express The phase vector and amplitude vector are obtained by the fast Fourier transform module. Then, the phase vector and the amplitude vector The inverse Fourier transform module will be calculated to reconstruct a multi-scale time domain feature map containing adjacent level features from a frequency domain perspective. The calculation process can be formally expressed as:

[0050]

[0051] in, It is an inverse Fourier transform module, which is used to restore the frequency domain spatial feature map back to the original time domain space. It is a feature upsampling module that uses a bilinear interpolation strategy to convert the amplitude vector Adjust to phase vector size for subsequent calculations. The multi-scale time domain feature map is reconstructed using the frequency domain information of the adjacent level feature map, and its size is the same as the input feature map. Consistent.

[0052] Furthermore, a multi-layer perceptron module is used to Perform feature learning, optimize the quality of the feature information of the frequency domain vector, and adjust its dimension to an appropriate size. The calculation process can be formally expressed as:

[0053]

[0054] in, , The distribution represents the multi-layer perceptron module that calculates different feature maps. It interactively calculates by giving each channel dynamic weight information to capture high-value channel information during the learning process. , Represents the frequency domain vector representation obtained after optimization.

[0055] Further, after the inverse Fourier transform module , Will be reconstructed into a time domain feature map. The calculation process can be formally expressed as:

[0056]

[0057] in, It is the feature map reconstructed after inverse Fourier transform, which contains multi-scale frequency domain feature information optimized by MLP module. Its size is similar to the input feature map. So far, the reconstruction of the two frequency domain feature maps is basically completed.

[0058] Furthermore, the feature map is obtained by frequency domain reconstruction Perform feature concatenation to obtain complete feature information. The calculation process can be formally expressed as:

[0059]

[0060] in, It represents the sum of the feature map information after frequency domain reconstruction, and its number of channels is the same as the input feature map twice as much.

[0061] Furthermore, the self-attention mechanism is used to capture global information under long-distance feature dependencies, and its calculation process can be formally expressed as:

[0062]

[0063] in, Represents the self-attention mechanism, Q represents the query matrix, K represents the key value, and V represents the value matrix. The three are obtained by linear mapping the input features, and the global features are obtained by establishing a long distance. Represents the feature map extracted by the self-attention mechanism.

[0064] Finally, point convolution is used to adjust the number of channels to ensure the dimension of the output feature map. The calculation process can be formally expressed as:

[0065]

[0066] in, Represents a point convolution module that adjusts the feature map in a learnable way The number of channels is used to facilitate subsequent calculations in the feature fusion network to obtain the final output At this point, the frequency domain feature reconstruction module is completed.

[0067] 2. Build a feature fusion network based on frequency domain feature reconstruction module optimization and its corresponding remote sensing image target detection model:

[0068] Most traditional target detection models based on convolutional neural networks use a backbone feature extraction network-feature fusion network-feature decoupling head structure as the basic architecture of the model. In the embodiment of the present invention, the remote sensing image target detection model based on frequency domain feature reconstruction follows this paradigm and introduces the feature fusion network optimized by the above-mentioned frequency domain feature reconstruction module to realize the extraction of multi-scale feature information, thereby improving the model's ability to utilize multi-scale feature information of typical remote sensing targets, helping the model to better capture multi-scale feature information about potential target areas, and realizing accurate remote sensing image target detection task recognition. The feature fusion network structure optimized based on the frequency domain feature reconstruction module is as follows: Figure 2 The following will describe in detail the construction method of the feature fusion network and remote sensing image target detection model of the present invention.

[0069] First, the remote sensing image and its label information are input into the model through the data loader. The calculation process of the data loader is expressed as:

[0070]

[0071] in, Indicates the size , input remote sensing image with 3 channels; Annotation information for the target in the image, including location and category information; It is a data loader function used to load data and enhance data;

[0072] Then the backbone feature extraction network is responsible for extracting features from the input image, and the convolution-based feature extraction network is responsible for extracting features from the input image. The computational process of feature extraction can be formally expressed as:

[0073]

[0074] in, Represents the feature map obtained after feature extraction, with a size of , and there are channels; It represents a backbone feature extraction network, which obtains feature information of the input remote sensing image by stacking convolution modules in layers. Typical backbone feature extraction networks include: ResNet, VGG, EfficientNet, etc. The method shown in this embodiment takes the basic convolution module as an example, but it should be noted that the remote sensing image target detection model based on frequency domain feature reconstruction of the present invention can be adapted to the general basic backbone feature extraction network through simple model replacement.

[0075] In a specific embodiment, the backbone feature extraction network It is composed of stacked convolution modules. Each basic convolution module contains a basic convolution operator (Conv), a normalization module (Batch Normalization) and an activation function (Relu). The basic convolution operator calculates the channel information of the input remote sensing image at a pixel-by-pixel level to achieve the purpose of feature extraction. The normalization module normalizes the calculated channel results to the standard range to facilitate subsequent calculations while avoiding abnormal gradient optimization problems. The activation function activates and inhibits the feature matrix of the potential target area to help the model distinguish between foreground and background information. The calculation process of the operation layer of each module can be formally expressed as:

[0076]

[0077] in, represents the basic convolution operation, is a normalization module, which keeps the number of channels it receives and outputs features unchanged. To input the original remote sensing image, It is the output feature map obtained after feature extraction. The activation function is responsible for activating the potential target area of ​​the output feature map. Definition It is a convolution module that includes the basic convolution operator, normalization module and activation function. Therefore, the feature extraction process of the deep neural network can be formally expressed as:

[0078]

[0079] in, , Indicates The convolutional module of the layer, is the input image, is the feature map obtained after the nth layer of convolution extraction. Based on this, we can obtain n feature maps of remote sensing images after deep feature extraction.

[0080] Furthermore, the feature fusion network is responsible for receiving the feature maps of the backbone feature extraction network for fusion to obtain the multi-scale characteristic information of the potential target.

[0081] In this embodiment, the basic model selects the feature map extracted by the backbone feature extraction network The feature fusion network is passed to the feature fusion network optimized based on the frequency domain feature reconstruction module for feature fusion; the calculation process of the feature fusion network is expressed as:

[0082]

[0083] in, It represents the feature fusion network optimized based on the frequency domain feature reconstruction module, which is used for feature fusion. It first receives the feature map of the backbone feature extraction network Then, the convolution operator and the frequency domain feature reconstruction module are used to achieve fusion to obtain the fused feature map ; The calculation process of the feature fusion network is expressed as:

[0084]

[0085] in, represents the feature concatenation operator, represents the convolution operator, represents the feature reshaping module, Represents the frequency domain feature reconstruction module.

[0086] Finally, the fused feature map is decoupled using the feature decoupling head. The feature decoupling network consists of a basic convolution module and a prediction function. The convolution module is used to receive the input remote sensing image features and process them in appropriate shapes and channel dimensions for subsequent calculations. The prediction function is used to predict the location of the potential target area of ​​the feature map, i.e., the category information, and output the prediction result. The calculation process of the detection head is expressed as:

[0087]

[0088] in, Classification information obtained for prediction; Location information obtained for prediction; A feature map outputted by the feature fusion network; is the classification function, is the rectangular box prediction function.

[0089] 3. Training and testing of remote sensing image target detection tasks

[0090] In one embodiment, after building a remote sensing image target detection model based on frequency domain feature reconstruction, focal loss, rotated IoU loss, and cross entropy loss are used to train the network for remote sensing image target detection tasks, which is formally expressed as:

[0091]

[0092] in, represents the target truth label, Represents the model prediction results. represents the total loss used in network training, Represents focal loss, which is used in the classification prediction branch of remote sensing image target detection tasks; Represents a spin box The loss function is used to calculate the distance between the predicted box and the target true value label corresponding to the input remote sensing image. When the network is trained until the loss no longer decreases, it means that the network training has reached stability, the training process is completed, and the trained remote sensing image target detection model based on frequency domain feature reconstruction is obtained.

[0093] Furthermore, the trained remote sensing image target detection model based on frequency domain feature reconstruction is used to test the remote sensing image to be tested, which is formally expressed as:

[0094]

[0095] in, represents the trained remote sensing image target detection model based on frequency domain feature reconstruction, and They respectively represent the remote sensing images to be tested and the corresponding target detection results.

[0096] From the description of the above embodiments, those skilled in the art can know that in order to overcome the insufficient ability of the feature extraction network to deal with the scale-changing targets, the present invention proposes an innovative and efficient remote sensing image target detection method. The method adopts a strategy based on frequency domain feature reconstruction, designs an efficient frequency domain feature reconstruction module, integrates it into the feature fusion network, obtains richer feature information from the frequency domain perspective, and effectively assists the feature fusion network to obtain the multi-scale information of the target. The feature fusion network optimized based on the frequency domain feature reconstruction module uses the frequency domain feature component representation of the feature map at different scale levels to reconstruct additional feature information to assist prediction. The reconstructed feature map supplements the dimension of the channel information under the original fusion path, helping the model to better capture the multi-scale feature information under the multi-dimensional perspective of the potential target area, and realize accurate remote sensing image target detection task recognition. This method significantly improves the target detection model's ability to perceive multi-dimensional information of scale-changing targets. Through frequency domain feature reconstruction, the model can more effectively capture richer multi-scale target feature information about the potential target area, thereby more accurately understanding the image content and correctly identifying and locating potential targets in remote sensing images.

[0097] The method of the present invention provides a more reliable and efficient solution for the field of remote sensing image target detection, and provides strong support for future remote sensing application fields. This method based on frequency domain feature reconstruction not only has broad application prospects in the fields of national defense security, environmental monitoring, urban planning, etc., but also provides a more feasible and accurate tool for remote sensing image interpretation, promoting the further development of remote sensing technology in practical applications.

[0098] Furthermore, the present invention also provides a remote sensing image target detection system based on frequency domain feature reconstruction, which is applied to a remote sensing image target detection method based on frequency domain feature reconstruction in the above embodiment to improve the performance of image target detection and facilitate accurate remote sensing image target detection task recognition; the system comprises:

[0099] Feature extraction network construction module, used to build a backbone feature extraction network assisted by deep principal component analysis;

[0100] A target detection model building module is used to construct a frequency domain feature reconstruction module, and to build a feature fusion network and a remote sensing image target detection model based on the frequency domain feature reconstruction module;

[0101] A target detection model training module is used to perform target detection task training on the remote sensing image target detection model;

[0102] The target detection model application module is used to use the trained remote sensing image target detection model to perform target detection on the remote sensing image to be tested.

[0103] The implementation principle and technical effects of the system provided in the embodiment of the present invention are the same as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the system embodiment, reference can be made to the corresponding contents in the aforementioned method embodiment, which will not be repeated here.

[0104] In addition, an embodiment of the present invention also provides a storage medium on which one or more programs readable by a computing device are stored, and the one or more programs include instructions. When the instructions are executed by the computing device, the computing device executes a remote sensing image target detection method based on frequency domain feature reconstruction in the above-mentioned embodiment.

[0105] In the embodiment of the present invention, the storage medium may be, for example, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples of storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, and any suitable combination thereof.

[0106] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems or computer program products, etc. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0107] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several distinct components, and by means of a suitably programmed computer.

[0108] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0109] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A remote sensing image target detection method based on frequency domain feature reconstruction, characterized in that: The method comprises the following steps: S1, construct frequency domain feature reconstruction module; S2. Building a feature fusion network and a remote sensing image target detection model based on the frequency domain feature reconstruction module; S3, performing target detection task training on the remote sensing image target detection model; S4, using the trained remote sensing image target detection model to perform target detection on the remote sensing image to be tested; In step S1, the frequency domain feature reconstruction module constructed includes: a Fourier transform module, an inverse Fourier transform module, a feature splicing module, a multi-layer perceptron module, an upsampling module, a self-attention module and a point convolution module, wherein: The Fourier transform module is used to map the feature map to the frequency domain space to extract the amplitude vector and the phase vector; the inverse Fourier transform module is used to restore the frequency domain space feature map back to the original time domain space; the multi-layer perceptron module is used to learn the effective feature information of the frequency domain vector and adjust its dimension to a suitable size; the upsampling module and the feature concatenation module are used to adjust the shape of the feature map; the self-attention module is used to establish long-distance feature dependencies to capture global information and optimize the information quality of the feature map; The upsampling module uses a bilinear interpolation method to adjust the amplitude vector to the size of the phase vector; In step S2, the remote sensing image target detection model constructed includes: a backbone feature extraction network, a feature fusion network optimized based on a frequency domain feature reconstruction module, and a feature decoupling head; The backbone feature extraction network is composed of a plurality of stacked convolution modules, each of which includes: a basic convolution operator, a normalization module and an activation function, wherein: The basic convolution operator calculates the channel information of the input remote sensing image at a pixel-by-pixel level; the normalization module normalizes the calculated channel results to the standard range; the activation function activates and inhibits the feature matrix of the potential target area; The feature fusion network optimized based on the frequency domain feature reconstruction module is used for feature fusion. It first receives the feature map of the backbone feature extraction network, and then uses the convolution operator and the frequency domain feature reconstruction module to achieve fusion to obtain the fused feature map; The feature decoupling head is used to decouple the fused feature map, and is composed of a basic convolution module and a prediction function; wherein the basic convolution module is used to receive the input remote sensing image features and process them in appropriate shapes and channel dimensions; the prediction function is used to predict the position of the potential target area of ​​the feature map and output the prediction result; In step S3, the remote sensing image target detection model is trained for target detection tasks using focal loss, rotation frame intersection-over-union loss, and cross entropy loss.

2. A remote sensing image target detection system based on frequency domain feature reconstruction, characterized in that: The method for remote sensing image target detection based on frequency domain feature reconstruction as claimed in claim 1 is applied to realize accurate remote sensing image target detection task recognition, and the system comprises: A target detection model building module is used to construct a frequency domain feature reconstruction module, and to build a feature fusion network and a remote sensing image target detection model based on the frequency domain feature reconstruction module; A target detection model training module is used to perform target detection task training on the remote sensing image target detection model; The target detection model application module is used to use the trained remote sensing image target detection model to perform target detection on the remote sensing image to be tested; The constructed frequency domain feature reconstruction module includes: a Fourier transform module, an inverse Fourier transform module, a feature splicing module, a multi-layer perceptron module, an upsampling module, a self-attention module and a point convolution module, wherein: The Fourier transform module is used to map the feature map to the frequency domain space to extract the amplitude vector and the phase vector; the inverse Fourier transform module is used to restore the frequency domain space feature map back to the original time domain space; the multi-layer perceptron module is used to learn the effective feature information of the frequency domain vector and adjust its dimension to a suitable size; the upsampling module and the feature concatenation module are used to adjust the shape of the feature map; the self-attention module is used to establish long-distance feature dependencies to capture global information and optimize the information quality of the feature map; The upsampling module uses a bilinear interpolation method to adjust the amplitude vector to the size of the phase vector; The remote sensing image target detection model constructed includes: a backbone feature extraction network, a feature fusion network optimized based on a frequency domain feature reconstruction module, and a feature decoupling head; The backbone feature extraction network is composed of a plurality of stacked convolution modules, each of which includes: a basic convolution operator, a normalization module and an activation function, wherein: The basic convolution operator calculates the channel information of the input remote sensing image at a pixel-by-pixel level; the normalization module normalizes the calculated channel results to the standard range; the activation function activates and inhibits the feature matrix of the potential target area; The feature fusion network optimized based on the frequency domain feature reconstruction module is used for feature fusion. It first receives the feature map of the backbone feature extraction network, and then uses the convolution operator and the frequency domain feature reconstruction module to achieve fusion to obtain the fused feature map; The feature decoupling head is used to decouple the fused feature map, and is composed of a basic convolution module and a prediction function; wherein the basic convolution module is used to receive the input remote sensing image features and process them in appropriate shapes and channel dimensions; the prediction function is used to predict the position of the potential target area of ​​the feature map and output the prediction result; Focus loss, rotation box intersection-over-union loss and cross entropy loss are used to train the remote sensing image target detection model for target detection tasks.