Dark space target enhancement detection method based on optical feature fusion

Abstract

The invention discloses a dark space target enhancement detection method based on optical feature fusion. The method comprises the following steps: acquiring multispectral data in a dark space scene; preprocessing the collected multispectral data, wherein the preprocessing comprises noise removal, image enhancement and feature extraction; performing feature fusion on the multispectral data by adopting an adaptive weight fusion algorithm to generate a fused multi-scale feature map; a deep learning target detection model is combined with an attention mechanism to carry out dark target detection and positioning on the fused feature map; a self-supervised learning method is introduced to train a deep learning target detection model, and generalization ability of the model is enhanced by using unlabeled data; and post-processing a detection result, and outputting position information and feature description of the dark space target. Through the multispectral feature fusion and deep learning detection technology, the problem that single optical imaging information is insufficient is solved, and the detection precision and robustness of a dark and weak target in a complex space environment are improved.

Description

Technical Field

[0001] The invention relates to a dark and weak space target enhanced detection method based on optical feature fusion, and belongs to the field of target detection. Background Art

[0002] In order to improve the detection capability of dim space targets, we should fully explore the inherent correlation characteristics of multiple types of perception information including multi-spectral imaging, medium and long wave infrared imaging, visible light imaging, laser imaging, etc., make up for the limitations of a single data source, and obtain more comprehensive and detailed target characteristic information, so as to enhance the detection capability of space targets.

[0003] In terms of feature fusion analysis, it is an important task to combine data from multi-source sensors and use computer vision, machine learning, multimodal analysis and other technologies to establish intelligent detection and autonomous analysis of space targets. However, in the face of the growing number of targets and unknown changes in space activities, it is difficult to obtain a sufficient set of learnable samples in a timely manner, which makes target detection and analysis more difficult. Therefore, it is urgent to establish a feature fusion enhanced detection technology for sparse sample space targets that is not limited to known samples and can integrate different optical perception features, so as to improve the ability of intelligent detection and intelligence analysis of space targets and provide strong technical support for space situational awareness. Summary of the invention

[0004] The purpose of the present invention is to provide a dark and weak space target enhanced detection method based on optical feature fusion, which can improve the dark and weak space target detection capability, make up for the limitations of a single data source, obtain more comprehensive and detailed target characteristic information, and enhance the ability of intelligent detection and intelligence analysis of space targets.

[0005] In order to achieve the above object, the present invention provides a dark space target enhanced detection method based on optical feature fusion, comprising the following steps: Step 1: Collect multispectral data in dark space scenes, including optical information such as mid- and long-wave infrared imaging, visible light imaging, and laser imaging; Step 2, preprocess the multispectral data collected in Step 1, including noise removal, image enhancement and feature extraction; Step 3, using the adaptive weight fusion algorithm to perform feature fusion on the multispectral data preprocessed in Step 2 to generate a fused multi-scale feature map; Step 4: Use the deep learning target detection model combined with the attention mechanism to detect and locate dim targets on the fused feature map; Step 5: Introduce self-supervised learning methods to train the deep learning target detection model and use unlabeled data to enhance the generalization ability of the model; Step 6: Post-process the detection results of Step 4, including background noise filtering and target precise positioning, and output the location information and feature description of the dark space target.

[0006] Furthermore, the visible light imaging collected in the Step 2 is subjected to Gaussian filtering for noise removal while retaining large-scale feature information. The infrared imaging and laser imaging collected in the Step 2 are subjected to wavelet transform denoising method, by decomposing the signal into frequency components of different scales, and then performing threshold processing on the high-frequency noise component, and then reconstructing the signal. The image is subjected to brightness and contrast enhancement processing using a histogram equalization method, and the grayscale histogram of the input image is transformed into a uniformly distributed histogram, thereby enhancing the visual effect of the image. The preprocessing steps are as follows: Step2.1-1, let the input original image be I, the Gaussian kernel size be k, calculate the weight value of the Gaussian kernel G, and the mathematical expression of the two-dimensional Gaussian function is: ; Where x and y represent spatial coordinates; σ is the standard deviation, which determines the distribution range of the Gaussian function, and G(x,y) is the weight value of the Gaussian kernel. The Gaussian filter uses the Gaussian function to generate the filter kernel. According to the above formula, a Gaussian filter kernel, where k represents the radius of the kernel, that is, k=3σ; Step 2.1-2: Fill the edges of the input image I (x, y) and then perform convolution operation on each pixel with the Gaussian kernel to generate a filtered image. , calculate the output value of each pixel , the formula is as follows: ; Where I(xi,yi) is the pixel value of the original image at position (xi,yi); G(x,y) is the weight of the position (x,y) in the Gaussian kernel; Step 2.1-3, in order to keep the image brightness unchanged, the Gaussian kernel needs to be normalized so that the sum of all weights is equal to 1. The normalization formula is as follows: ; Step 2.2-1, use the wavelet transform denoising method to denoise the collected infrared imaging and laser imaging. Specifically, decompose the image into different scales and frequency ranges. The noise is mainly distributed in the high-frequency part. By thresholding the high-frequency noise coefficient, retain the important feature information of the image and remove the noise. Perform L-layer wavelet decomposition on the image f (x, y), and the result is: ; Wavelet decomposition uses a low-pass filter h(n) and a high-pass filter g(n) to complete the decomposition through a convolution operation: ; Among them A L is the low-frequency coefficient of the Lth layer, D j is the high frequency coefficient of the jth layer, A j+1 and D j+1 They represent the low-frequency and high-frequency components of the j+1th layer respectively, and 2k represents downsampling.

[0007] Step 2.2-2, use the soft threshold method to j Denoising, specifically, by smoothing the high-frequency coefficients, effectively removes noise while retaining the main features of the image. The main operation is to subtract the threshold T from the absolute value of the high-frequency coefficient, and then restore the direction of the coefficient according to the sign. The mathematical expression used is: ; Step 2.2-3, the selection of threshold T has a great influence on the denoising effect, so the optimal threshold is determined by the global threshold, the formula is: ; Where N is the number of pixels in the image, is the standard deviation of the noise, which can be estimated by the median of the high-frequency coefficients, expressed as .

[0008] Step 2.2-4, the high frequency components processed in Step 2.2-2 and the low-frequency component A in Step 2.2-1 L Perform inverse wavelet transform to reconstruct the denoised image. The inverse wavelet transform formula is as follows: ; The inverse transform process is the opposite of the decomposition, combining a low-pass filter h(n) and a high-pass filter g(n) for upsampling and convolution operations.

[0009] Step 2.3-1, use the histogram equalization method to enhance the brightness and contrast of the image. Specifically, by redistributing the grayscale values ​​of the image pixels, the brightness and contrast of the image are improved. The basic idea is to transform the grayscale histogram of the input image into a uniformly distributed histogram, thereby enhancing the visual effect of the image and counting the number of pixels at each gray level in the input image. Then calculate the gray level The probability density of , the calculation expression is as follows: ; in Indicates the gray value is The number of pixels, Indicates the total number of pixels in the image.

[0010] Step 2.3-2, for each gray level , calculate its cumulative distribution function Represents the cumulative probability from gray level 0 to k, and the expression is as follows: ; Step2.3-3, according to the cumulative distribution function, the input gray value Map to the new output grayscale , the grayscale values ​​of all pixels in the input image Replace with the corresponding output grayscale value , generate the histogram equalized image g(x,y), the expression is as follows: ; Among them, L-1 is the grayscale value range of the output image, and round() rounds the result to the nearest integer.

[0011] Furthermore, the specific steps of Step 3 are: The adaptive weight fusion algorithm optimizes the fusion effect by dynamically calculating the weight of each input data (or feature). Its main purpose is to assign weights according to the importance of the data (such as energy, confidence or attention, etc.) and achieve efficient fusion of information. First, define the i-th spectral data feature map Contribution weight , weight For each position (h, w), the calculation is performed adaptively according to the local contribution based on the attention mechanism, which is expressed by the following formula: ; Where β is the temperature parameter, which is used to adjust the smoothness of the weight distribution; Secondly, according to the adaptive weight For all feature maps Perform weighted summation to generate the fused feature map fusion formula: ; in, is the input multi-spectral feature map, is the dynamically calculated local adaptive weight; Then, convolutional layers with different receptive fields are used to fuse features at different levels, strengthening the deep network's ability to focus on dim targets. While obtaining the semantic feature expression of dim targets, the accuracy of target positioning is taken into account to improve the detection effect. Through multiple convolutions and downsampling, feature maps of different resolutions are generated. The expression is as follows: ; Among them is The feature map of the lth layer, conv represents the convolution operation, k is the convolution kernel size, s is the step size, and L is the number of multi-scale layers.

[0012] Furthermore, the step of using the deep learning target detection model combined with the attention mechanism to detect and locate dim targets on the fused feature map in Step 4 is: First, the attention mechanism is used to enhance the feature map. Specifically, the core idea of ​​the attention mechanism is to enhance the performance of important areas or features in the image while suppressing unimportant parts. Global pooling is performed on each channel to obtain the global description vector of the channel. The formula for global average pooling is: ; Then, the channel weights are generated through two layers of fully connected networks and activation functions. The weight calculation formula is as follows: ; in and is the weight matrix, is the ReLU activation function, is the Sigmoid function.

[0013] Finally, the channel weight Acting on the feature map, the formula is: ; Furthermore, the specific steps of Step 5 are: By contrasting the loss function, the similarity between positive samples is maximized and the similarity between negative samples is minimized. The formula is as follows: ; Where T is the temperature parameter, M is the total number of samples in the batch, and z 1 and z 2 is the feature representation of the two views, sim(z 1 ,z 2 ) is the cosine similarity between feature representations, z j is the feature representation of other samples.

[0014] Furthermore, the steps of filtering the background noise and accurately locating the target in the detection result in Step 6 are as follows: Use the NMS algorithm to filter redundant detection frames and retain the optimal frame. Specifically, NMS is mainly used to filter out low-confidence frames that overlap with other frames, thereby improving the accuracy of target detection. First, sort the candidate frames in descending order according to the confidence score, select the detection frame with the highest confidence score Bmax in turn, and calculate the confidence score with the remaining frames B jThe IoU between them will remove the detection boxes whose IoU with Bmax is greater than the threshold until all boxes are processed. The calculation formula is as follows: ; Through the bounding box regression branch of the target detection model, the positioning accuracy of the target is further optimized, and the optimization goal is to minimize the regression error.

[0015] Beneficial effects: Through multi-spectral feature fusion and deep learning detection technology, the problem of insufficient single optical imaging information is effectively solved, and the detection accuracy and robustness of dim targets in complex space environments are improved; it can improve the detection capability of dim space targets, make up for the limitations of a single data source, obtain more comprehensive and detailed target characteristic information, and enhance the ability of intelligent detection and intelligence analysis of space targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is a principle diagram of the present invention using wavelet transform for denoising; Figure 3 It is a schematic diagram of an adaptive feature processing module for generating a fused multi-scale feature map by using an adaptive weight fusion algorithm to perform feature fusion on pre-processed multi-spectral data; Figure 4 It is an architecture diagram of the present invention using a self-supervised learning method to train a detection model. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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. 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.

[0019] like Figure 1 As shown, a dark and weak space target enhanced detection method based on optical feature fusion includes the following steps: Step 1: Collect multispectral data in dark space scenes, including optical information such as mid- and long-wave infrared imaging, visible light imaging, and laser imaging; Step 2, preprocess the multispectral data collected in Step 1, including noise removal, image enhancement and feature extraction; Step 3, using the adaptive weight fusion algorithm to perform feature fusion on the multi-spectral data preprocessed in Step 2 to generate a fused multi-scale feature map; Step 4: Use the deep learning target detection model combined with the attention mechanism to detect and locate dim targets on the fused feature map; Step 5: Introduce self-supervised learning methods to train the deep learning target detection model and use unlabeled data to enhance the generalization ability of the model; Step 6: Post-process the detection results of Step 4, including background noise filtering and target precise positioning, and output the location information and feature description of the dark space target.

[0020] As a preferred implementation, the visible light imaging collected in Step 2 is subjected to Gaussian filtering for noise removal while retaining large-scale feature information. The infrared imaging and laser imaging collected in Step 2 are subjected to wavelet transform denoising method, by decomposing the signal into frequency components of different scales, then performing threshold processing on the high-frequency noise components, and then reconstructing the signal. The principle diagram of wavelet transform denoising is as follows: Figure 2 As shown, the histogram equalization method is used to enhance the brightness and contrast of the image, and the grayscale histogram of the input image is transformed into a uniformly distributed histogram, thereby enhancing the preprocessing of the visual effect of the image; Step2.1-1, let the input original image be I, the Gaussian kernel size be k, calculate the weight value of the Gaussian kernel G, and the mathematical expression of the two-dimensional Gaussian function is: ; Where x and y represent spatial coordinates; σ is the standard deviation, which determines the distribution range of the Gaussian function, and G(x,y) is the weight value of the Gaussian kernel. The Gaussian filter uses the Gaussian function to generate the filter kernel. According to the above formula, a Gaussian filter kernel, where k represents the radius of the kernel, that is, k=3σ; Step 2.1-2: Fill the edges of the input image I (x, y) and then perform convolution operation on each pixel with the Gaussian kernel to generate a filtered image. , calculate the output value of each pixel , the formula is as follows: ; Where I(xi,yi) is the pixel value of the original image at position (xi,yi); G(x,y) is the weight of the position (x,y) in the Gaussian kernel; Step 2.1-3, in order to keep the image brightness unchanged, the Gaussian kernel needs to be normalized so that the sum of all weights is equal to 1. The normalization formula is as follows: ; Step 2.2-1, use the wavelet transform denoising method to denoise the collected infrared imaging and laser imaging. Specifically, decompose the image into different scales and frequency ranges. The noise is mainly distributed in the high-frequency part. By thresholding the high-frequency noise coefficient, retain the important feature information of the image and remove the noise. Perform L-layer wavelet decomposition on the image f (x, y), and the result is: ; Wavelet decomposition uses a low-pass filter h(n) and a high-pass filter g(n) to complete the decomposition through a convolution operation: ; Among them A L is the low-frequency coefficient of the Lth layer, D j is the high frequency coefficient of the jth layer, A j+1 and D j+1 They represent the low-frequency and high-frequency components of the j+1th layer respectively, and 2k represents downsampling.

[0021] Step 2.2-2, use the soft threshold method to j Denoising, specifically, by smoothing the high-frequency coefficients, effectively removes noise while retaining the main features of the image. The main operation is to subtract the threshold T from the absolute value of the high-frequency coefficient, and then restore the direction of the coefficient according to the sign. The mathematical expression used is: ; Step 2.2-3, the selection of threshold T has a great influence on the denoising effect, so the optimal threshold is determined by the global threshold, the formula is: ; Where N is the number of pixels in the image, is the standard deviation of the noise, which can be estimated by the median of the high-frequency coefficients, expressed as ; Step 2.2-4, the high frequency components processed in Step 2.2-2 and the low-frequency component A in Step 2.2-1 L Perform inverse wavelet transform to reconstruct the denoised image. The inverse wavelet transform formula is as follows: ; The inverse transform process is the opposite of the decomposition, combining the low-pass filter h(n) and the high-pass filter g(n) for upsampling and convolution operations; Step 2.3-1, use the histogram equalization method to enhance the brightness and contrast of the image. Specifically, by redistributing the grayscale values ​​of the image pixels, the brightness and contrast of the image are improved; the basic idea is to transform the grayscale histogram of the input image into a uniformly distributed histogram, thereby enhancing the visual effect of the image and counting the number of pixels at each gray level in the input image. Then calculate the gray level The probability density of , the calculation expression is as follows: ; in Indicates the gray value is The number of pixels, Indicates the total number of pixels in the image; Step 2.3-2, for each gray level , calculate its cumulative distribution function Represents the cumulative probability from gray level 0 to k, and the expression is as follows: ; Step2.3-3, according to the cumulative distribution function, the input gray value Map to the new output grayscale , the grayscale values ​​of all pixels in the input image Replace with the corresponding output grayscale value , generate the histogram equalized image g(x,y), the expression is as follows: ; Among them, L-1 is the grayscale value range of the output image, and round() rounds the result to the nearest integer.

[0022] As a preferred implementation, the adaptive weight fusion algorithm optimizes the fusion effect by dynamically calculating the weight of each input data. Its main purpose is to assign weights according to the importance of the data and achieve efficient fusion of information, such as Figure 3 The specific steps of Step 3 are: First, define the i-th spectral data feature map Contribution weight , weight For each position (h, w), the calculation is performed adaptively according to the local contribution based on the attention mechanism, which is expressed by the following formula: ; Where β is the temperature parameter, which is used to adjust the smoothness of the weight distribution; Secondly, according to the adaptive weight For all feature maps Perform weighted summation to generate the fused feature map fusion formula: ; in, is the input multi-spectral feature map, is the dynamically calculated local adaptive weight; Then, convolutional layers with different receptive fields are used to fuse features at different levels, strengthening the deep network's ability to focus on dim targets. While obtaining the semantic feature expression of dim targets, the accuracy of target positioning is taken into account to improve the detection effect. Through multiple convolutions and downsampling, feature maps of different resolutions are generated. The expression is as follows: ; Among them is The feature map of the lth layer, conv represents the convolution operation, k is the convolution kernel size, s is the step size, and L is the number of multi-scale layers.

[0023] As a preferred implementation, the specific steps of Step 4 are: First, the attention mechanism is used to enhance the feature map. Specifically, the core idea of ​​the attention mechanism is to enhance the performance of important areas or features in the image by focusing on these areas, while suppressing unimportant parts, and performing global pooling on each channel to obtain the global description vector of the channel. The formula for global average pooling is: ; Then, the channel weights are generated through two layers of fully connected networks and activation functions. The weight calculation formula is as follows: ; in and is the weight matrix, is the ReLU activation function, is the Sigmoid function; Finally, the channel weight Acting on the feature map, the formula is: ; As a preferred implementation, the specific steps of Step 5 are: By contrasting the loss function, the similarity between positive samples is maximized and the similarity between negative samples is minimized. The formula is as follows: ; Where T is the temperature parameter, M is the total number of samples in the batch, and z 1 and z 2 is the feature representation of the two views, sim(z 1 ,z2 ) is the cosine similarity between feature representations, z j is the feature representation of other samples.

[0024] As a preferred implementation, the specific steps of Step 6 are: Use the NMS algorithm to filter redundant detection frames and retain the optimal frame. Specifically, NMS is mainly used to filter out low-confidence frames that overlap with other frames, thereby improving the accuracy of target detection. First, the candidate frames are sorted in descending order by confidence score, and the detection frame with the highest confidence score Bmax is selected in turn, and the confidence interval between the candidate frames Bmax and the remaining frames Bmax is calculated. j The IoU between them will remove the detection boxes whose IoU with Bmax is greater than the threshold until all boxes are processed. The calculation formula is as follows: ; Through the bounding box regression branch of the target detection model, the positioning accuracy of the target is further optimized, and the optimization goal is to minimize the regression error.

[0025] In summary, the enhanced detection method of dim space targets based on optical feature fusion can improve the detection capability of dim space targets, make up for the limitations of a single data source, obtain more comprehensive and detailed target characteristic information, and enhance the ability of intelligent detection and intelligence analysis of space targets.

[0026] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for enhanced detection of dark and weak space targets based on optical feature fusion, characterized in that: The steps include: Step 1: Collect multispectral data in dark space scenes, including mid- and long-wave infrared imaging, visible light imaging, and laser imaging optical information; Step 2, preprocess the multispectral data collected in Step 1, including noise removal, image enhancement and feature extraction; Step 3, using the adaptive weight fusion algorithm to perform feature fusion on the multi-spectral data preprocessed in Step 2 to generate a fused multi-scale feature map; Step 4: Use the deep learning target detection model combined with the attention mechanism to detect and locate dim targets on the fused feature map; Step 5: Introduce self-supervised learning methods to train the deep learning target detection model and use unlabeled data to enhance the generalization ability of the model; Step 6: Post-process the detection results of Step 4, including background noise filtering and target precise positioning, and output the location information and feature description of the dark space target.

2. The method for enhanced detection of dark and weak space targets based on optical feature fusion according to claim 1 is characterized in that: The specific steps of Step 2 are: Step2.1-1, let the input original image be I, the Gaussian kernel size be k, calculate the weight value of the Gaussian kernel G, and the mathematical expression of the two-dimensional Gaussian function is: ； Where x, y represent spatial coordinates; σ is the standard deviation, G(x, y) is the weight value of the Gaussian kernel; the Gaussian filter uses the Gaussian function to generate the filter kernel. According to the above formula, a Gaussian filter kernel, where k represents the radius of the kernel, that is, k=3σ; Step 2.1-2: Fill the edges of the input image I (x, y) and then perform convolution on each pixel with the Gaussian kernel to generate a filtered image. , calculate the output value of each pixel The formula is as follows: ； Where I(xi,yi) is the pixel value of the original image at position (xi,yi); G(x,y) is the weight of the position (x,y) in the Gaussian kernel; Step 2.1-3, the Gaussian kernel needs to be normalized so that the sum of all weights is equal to 1. The normalization formula is as follows: ； Step 2.2-1, use the wavelet transform denoising method to denoise the collected infrared imaging and laser imaging, and perform L-layer wavelet decomposition on the image f (x, y), the result is: ； Wavelet decomposition uses a low-pass filter h(n) and a high-pass filter g(n) to complete the decomposition through a convolution operation: ； Among them A L is the low-frequency coefficient of the Lth layer, D j is the high frequency coefficient of the jth layer, A j+1 and D j+1 They represent the low-frequency and high-frequency components of the j+1th layer, respectively, and 2k represents downsampling; Step 2.2-2, for high frequency component D j The main operation of denoising by the soft threshold method is to subtract the threshold T from the absolute value of the high-frequency coefficient, and then restore the direction of the coefficient according to the sign. The mathematical expression used is: ； Step 2.2-3, determine the optimal threshold through the global threshold, the formula is: ； Where N is the number of pixels in the image, is the standard deviation of the noise, which can be estimated by the median of the high-frequency coefficients. The expression is: ； Step 2.2-4, the high frequency components processed in Step 2.2-2 and the low-frequency component A in Step 2.2-1 L Perform inverse wavelet transform to reconstruct the denoised image. The inverse wavelet transform formula is as follows: ； The inverse transform process is the opposite of the decomposition, combining the low-pass filter h(n) and the high-pass filter g(n) for upsampling and convolution operations; Step 2.3-1. Use the histogram equalization method to enhance the brightness and contrast of the image and count the number of pixels at each gray level in the input image. Then calculate the gray level The probability density of , the calculation expression is as follows: ； in Indicates the gray value is The number of pixels, Indicates the total number of pixels in the image; Step 2.3-2, for each gray level , calculate its cumulative distribution function Represents the cumulative probability from gray level 0 to k, and the expression is as follows: ； Step2.3-3, according to the cumulative distribution function, the input gray value Map to the new output grayscale , the grayscale values ​​of all pixels in the input image Replace with the corresponding output grayscale value , generate the histogram equalized image g(x,y), the expression is as follows: 。 3. The method for enhanced detection of dark and weak space targets based on optical feature fusion according to claim 1 is characterized in that: The specific steps of Step 3 are: First, define the i-th spectral data feature map Contribution weight , the weights are adaptively calculated based on the local contribution of the attention mechanism, expressed as follows: ； Where β is the temperature parameter, which is used to adjust the smoothness of the weight distribution; Then according to the adaptive weight For all feature maps Perform weighted summation to generate the fused feature map fusion formula: ； in, is the input multi-spectral feature map, is a dynamically calculated local adaptive weight.

4. The method for enhanced detection of dark and weak space targets based on optical feature fusion according to claim 3 is characterized in that: The specific steps of Step 4 are: First, the feature map is enhanced using the attention mechanism, and global pooling is performed on each channel to obtain the global description vector of the channel. The formula for global average pooling is: ； Then, the channel weights are generated through two layers of fully connected networks and activation functions. The weight calculation formula is as follows: ； in and is the weight matrix, is the ReLU activation function, is the Sigmoid function; Finally, the channel weight Acting on the feature map, the formula is: 。 5. The method for enhanced detection of dark and weak space targets based on optical feature fusion according to claim 1, characterized in that: The specific steps of Step 5 are: By contrasting the loss function, the similarity between positive samples is maximized and the similarity between negative samples is minimized. The formula is as follows: ； Where T is the temperature parameter, M is the total number of samples in the batch, z1 and z2 are the feature representations of the two views, sim(z1,z2) is the cosine similarity between the feature representations, and z j is the feature representation of other samples.

6. The method for enhanced detection of dark and weak space targets based on optical feature fusion according to claim 1, characterized in that: The specific steps of Step 6 are: Use the NMS algorithm to filter redundant detection frames, retain the optimal frame, sort the candidate frames in descending order by confidence score, and select the detection frame with the highest confidence in turn. max , calculate and the remaining box B j The IoU between max The detection frames whose IoU is greater than the threshold are removed until all frames are processed. The calculation formula is as follows: ； Through the bounding box regression branch of the target detection model, the positioning accuracy of the target is further optimized, and the optimization goal is to minimize the regression error.

Images