Self-Supervised Learning-Based Contrast Feature Enhancement Method for Low-Illumination Space Targets

Abstract

A method for enhancing contrast features of low-illuminance space targets based on self-supervised learning, the steps include: collecting space images; performing preprocessing operations on the collected space images; extracting features from the data after data augmentation; suppressing noise on the images after feature extraction; enhancing features of key regions of the images after noise suppression; classifying similar features of the images after feature enhancement and performing target recognition and analysis; enhancing the images after the target analysis has been completed. The present invention realizes image feature enhancement driven by self-supervised learning and obtains more comprehensive and accurate scene information; through the scale-invariant feature transform algorithm, it ensures the stability and consistency of image features under low-illumination conditions and improves the accuracy of feature enhancement. It effectively suppresses noise interference in low-illumination environments, enhances the saliency of target features, and expands the application scope of the low-illumination image feature enhancement technology based on self-supervised learning.

Description

Technical Field

[0001] The present invention relates to a method for enhancing contrast features of space targets in low - illumination spaces based on self - supervised learning, belonging to the technical field of image processing. Background Art

[0002] In low - illumination environments, the detection and recognition of space targets are an important research direction in the fields of image processing and computer vision. Due to insufficient illumination, low - illumination images are usually accompanied by problems such as high noise levels, low contrast, and detail loss. These factors seriously affect the accurate detection and recognition of space targets.

[0003] Traditional low - illumination image enhancement methods mainly rely on image enhancement techniques such as histogram equalization, Retinex theory and its variants. These methods attempt to restore the details and structure of the image by adjusting the brightness and contrast of the image. However, traditional enhancement techniques have the following main problems when dealing with low - illumination images: Insufficient noise suppression: In low - illumination environments, the noise level in the image increases significantly. Traditional methods usually use fixed filters or noise models for denoising. However, this method lacks adaptability and is difficult to effectively suppress noise under different noise types and intensities, resulting in the loss of target details or noise residue; Limited contrast adjustment: Although methods such as histogram equalization can improve the contrast of the image, under low - illumination conditions, excessive contrast enhancement may lead to artifacts and edge distortion in the image, affecting the accurate recognition of space targets; Insufficient feature extraction: Traditional methods mainly focus on adjusting the brightness and contrast of the image, lacking effective extraction and optimization of the deep features of the image. This results in the features of space targets not being significant enough in complex backgrounds and low - signal - to - noise ratio environments, making it difficult to achieve high - precision detection and recognition; Dependence on manual parameter adjustment: Many traditional enhancement methods require manual setting of parameters, such as the size of the filter, the intensity of contrast adjustment, etc. This not only increases the complexity of use but also limits the generalization ability of the method in different scenarios.

[0004] In addition, in practical applications such as military reconnaissance, night - time surveillance, and traffic surveillance under low light levels, the real - time processing and high - quality enhancement of low - illumination images pose higher requirements for the performance of the system. Traditional methods often struggle to balance processing efficiency and image quality in these high - demand scenarios and cannot meet the requirements of real - time and high - resolution.

[0005] To address the above problems, there is an urgent need for a method that can effectively enhance the contrast features of images under low - illumination conditions, which can not only adaptively suppress noise, optimize image features, but also automatically adapt to different environmental conditions without manual parameter adjustment, improving the detection accuracy and robustness of space targets. Summary of the Invention

[0006] The object of the present invention is to provide a method for enhancing contrast features of low - illumination space targets based on self - supervised learning. Through a self - supervised contrast learning mechanism and an adaptive noise suppression technique, efficient enhancement of low - illumination images and precise detection of space targets are achieved, overcoming the deficiencies of traditional methods in noise suppression, contrast adjustment, and feature extraction, and meeting the performance requirements in high - demand application scenarios.

[0007] To achieve the above object, the present invention provides a method for enhancing contrast features of low - illumination space targets based on self - supervised learning, including the following steps:

[0008] Step1. Collect space images in a low - illumination space scene;

[0009] Step2. Perform data input and data augmentation operations on the low - illumination space images collected in Step1, including rotation, cropping, etc.;

[0010] Step3. Use a self - supervised contrast learning mechanism to extract features from the images after data augmentation in Step2;

[0011] Step4. Use an adaptive noise suppression mechanism to perform dynamic noise suppression on the images after feature extraction in Step3;

[0012] Step5. Use an attention mechanism to enhance the features of the key regions of the images after dynamic noise suppression in Step4;

[0013] Step6. Use a density - based clustering algorithm to classify similar features of the images after feature enhancement in Step5 and perform target recognition and analysis;

[0014] Step7. Use a non - regularized self - supervised learning mechanism to enhance the images after target analysis in Step6.

[0015] Further, the specific steps of Step2 are as follows:

[0016] Step2.1. Perform geometric transformations on the faint space images, including rotation, translation, cropping, scaling, and flipping;

[0017] Step2.2. Perform color - space transformations on the faint space images after geometric transformations, including brightness adjustment, contrast adjustment, color perturbation, and grayscaling, and use these enhanced samples as "positive samples", while using samples from different images as "negative samples" for comparison.

[0018] Further, the specific steps of Step3 are as follows:

[0019] Take the enhanced data obtained after Step 2.2 as "positive samples", and at the same time take samples from different images as "negative samples" for comparison. Optimize the feature representation through the following formula for contrastive learning objectives:

[0020] ;

[0021] where, is the image feature representation extracted by the neural network, is the temperature parameter, is the negative sample, and N is the total number of negative samples. The objective of this loss function is to make the feature vectors between positive samples as similar as possible, while the feature vectors of different targets are far from each other.

[0022] Furthermore, the specific steps of Step 4 are as follows:

[0023] Step 4.1: First, estimate the noise intensity in the local area of the image. Using the self-supervised learning framework, the model can automatically identify which areas in the image may contain noise, such as areas with unclear texture and blurred edges, and estimate and suppress the noise based on the noise distribution model of these areas; Secondly, introduce the noise weight map W to represent the noise intensity of each area. In this way, the noise area will obtain a higher weight for suppression, while the target area will retain more details. The noise suppression loss function is as shown in the formula:

[0024] ;

[0025] where, represents the feature of the i-th area in the input faint space image S, is the feature of the corresponding area in the denoised image , is the adaptive noise weight adjusted according to the regional noise intensity, is the indicator function. When belongs to the noise area, the weight of this area increases.

[0026] Step4.2. Further introduce a self-supervised contrastive learning mechanism to perform feature learning and contrast optimization on the image after dynamic noise suppression, so as to improve the robustness and accuracy of the target features. In the contrastive learning framework of self-supervised learning, by maximizing the similarity between different perspectives and augmented views, the model can automatically learn the key features of spatial targets without labeled data and automatically suppress the noise regions. During the contrastive learning process, design a mechanism that includes positive and negative sample contrast. The positive samples are generated through image augmentation, and the negative samples come from different images or background regions. The optimization goal is to make the representation of the target region in the feature space more compact while distinguishing the features of the background region from those of the target. This optimization process can effectively reduce the impact of noise on the target features, thereby improving the detection performance of the target. The loss function of contrastive learning is as shown in the formula:

[0027] ;

[0028] where, and are the feature representations from the input image S and the augmented image T respectively, is the temperature coefficient, is the negative sample, and N is the number of negative samples.

[0029] Furthermore, the step of using the attention mechanism to enhance the key regions of the image after dynamic noise suppression in Step5 is as follows:

[0030] Use the attention mechanism to enhance the key regions (such as the edges and core parts of the target) with the latent features of the image obtained after Step4.2. During the training process, the model will automatically learn which regions are most important for target recognition, thereby enhancing the feature expression ability of these regions. The attention mechanism can be implemented through the following formula:

[0031] ;

[0032] where, is the learned attention weight, is the feature representation of the i-th local region.

[0033] Furthermore, the step of using the density-based clustering algorithm method to classify similar features and perform target recognition and analysis on the image after feature enhancement in Step6 is as follows:

[0034] Use the density-based clustering algorithm to group similar features into the same class, thereby identifying and analyzing new spatial targets. The density-based clustering algorithm can be implemented through the following formula:

[0035] ;

[0036] Among them, is the j-th clustering center, is the sample set of the j-th cluster, is the sample point in the feature representation.

[0037] Furthermore, the steps of using the non-regularized self-supervised learning mechanism to perform image enhancement on the image after target analysis in Step6 in Step7 are as follows:

[0038] Step7.1. Use the bilateral filtering technique to smooth the input low-illumination image and preliminarily estimate the illumination map, which is achieved through the following formula:

[0039] ;

[0040] Among them, represents the preliminary illumination map, and S represents the input low-illumination image:

[0041] Step7.2. Use the self-supervised learning model to further optimize the just-obtained preliminary illumination map to obtain a more accurate and consistent optimized illumination map, which can be achieved through the following formula:

[0042] ;

[0043] Among them, are the model parameters, obtained through optimization training; is the self-supervised learning model; is the preliminary illumination map, and S is the input low-illumination image.

[0044] Step7.3. Apply the correction factor r to adjust the illumination map to obtain the adjusted illumination map, which can be achieved through the following formula:

[0045] ;

[0046] Among them, is the reflectivity map, is the input low-illumination image, is the adjusted illumination map;

[0047] Take the reflectivity map R as the enhanced image and achieve visualization by amplifying its intensity value, which can be achieved through the following formula:

[0048] ;

[0049] Among them, The function is used to amplify the intensity value of the reflectivity map to make the obtained image brighter and clearer.

[0050] Beneficial effects: This method can effectively enhance the target features in images under low illumination conditions, improving the accuracy and robustness of target detection. By combining a self-supervised contrast learning mechanism and an adaptive noise suppression technique, the present invention achieves efficient processing of low-illumination images, overcoming the deficiencies of traditional methods in noise suppression, contrast adjustment, and feature extraction. This method does not rely on a large amount of labeled data, has good generalization ability and adaptability, and is applicable to high-demand application scenarios such as military reconnaissance, night monitoring, and traffic monitoring, significantly improving the detection and recognition effect of spatial targets in low-illumination environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0052] Figure 1 is a flowchart of the present invention;

[0053] Figure 2 is a schematic diagram of feature extraction based on self-supervised contrast learning;

[0054] Figure 3 is a schematic diagram of optimizing the spatial target features by adaptive noise suppression;

[0055] Figure 4 is a schematic diagram of estimating the illumination mapping by non-regularized self-supervised learning. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all of them. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0057] Taking a low-illumination urban street as an example, as Figure 1 shown, the method for enhancing the contrast features of spatial targets with low illumination based on self-supervised learning includes the following steps:

[0058] Step1. Collect spatial images in a low-illumination spatial scene;

[0059] Step2. Perform data input and data augmentation operations on the low-illumination spatial images collected in Step1, including rotation, cropping, etc.;

[0060] Step 3. Extract features from the images after data augmentation in Step 2 using a self-supervised contrastive learning mechanism;

[0061] Step 4. Perform dynamic noise suppression on the images after feature extraction in Step 3 using an adaptive noise suppression mechanism;

[0062] Step 5. Enhance the features of the key regions of the images after dynamic noise suppression in Step 4 using an attention mechanism;

[0063] Step 6. Use a density-based clustering algorithm to classify similar features of the images after feature enhancement in Step 5 and perform object recognition and analysis;

[0064] Step 7. Perform image enhancement on the images after object analysis in Step 6 using an unregularized self-supervised learning mechanism.

[0065] As a preferred implementation manner, the specific steps of Step 2 are:

[0066] Step 2.1. Perform geometric transformations on the faint space images, including rotation, translation, cropping, scaling, and flipping;

[0067] Step 2.2. Perform color space transformations on the faint space images after geometric transformations, including brightness adjustment, contrast adjustment, color perturbation, and grayscaling, and use these enhanced samples as "positive samples", while using samples from different images as "negative samples" for comparison.

[0068] Furthermore, the steps of using a self-supervised contrastive learning mechanism to extract features from the images after data augmentation in Step 2 in Step 3 are as follows:

[0069] Use the enhanced data obtained after Step 2.2 as "positive samples", while using samples from different images as "negative samples" for comparison, and optimize the feature representation through the following formula for contrastive learning objectives:

[0070] ;

[0071] where is the image feature representation extracted by the neural network, is the temperature parameter, is the negative sample, N is the total number of negative samples, and the objective of this loss function is to make the feature vectors between positive samples as similar as possible, while the feature vectors of different targets are far from each other, as shown in Figure 2As shown, in the experiment, after image enhancement, the contrast and brightness are significantly improved: the picture contrast is increased by 25.43%, and the brightness is improved by 15.25 dB.

[0072] As a preferred implementation, the specific steps of Step 4 are as follows:

[0073] Step 4.1: First, estimate the noise intensity in the local area of the image. Using the self-supervised learning framework, the model can automatically identify which areas in the image may contain noise, such as areas with unclear texture and blurred edges, and estimate and suppress the noise based on the noise distribution model of these areas. Secondly, introduce a noise weight map W to represent the noise intensity of each area. In this way, the noise area will obtain a higher weight for suppression, while the target area retains more details. The noise suppression loss function is as shown in the formula:

[0074] ;

[0075] where, represents the feature of the i-th area in the input dim image S, is the feature of the corresponding area in the denoised image is the adaptive noise weight adjusted according to the regional noise intensity, is an indicator function. When belongs to the noise area, the weight of this area increases. Experiments show that after self-supervised contrast learning, the target recognition accuracy and feature extraction quality have been greatly improved: the target recognition accuracy has been increased to 88.34%, and the feature similarity has been increased by 17.89%;

[0076] Step 4.2: Further introduce a self-supervised contrast learning mechanism to perform feature learning and contrast optimization on the image after dynamic noise suppression, so as to improve the robustness and accuracy of the target features. The contrast learning framework in self-supervised learning maximizes the similarity between different perspectives and enhanced views, enabling the model to automatically learn the key features of spatial targets without labeled data and automatically suppress the noise areas. During the contrast learning process, design a mechanism including positive and negative sample contrast. The positive samples are generated by image enhancement, and the negative samples come from different images or background areas. The optimization goal is to make the representation of the target area in the feature space more compact, while separating the features of the background area from the target. This optimization process can effectively reduce the impact of noise on the target features, thereby improving the target detection performance. The loss function of contrast learning is as shown in the formula:

[0077] ;

[0078] where, and They are the feature representations from the input image S and the enhanced image T respectively, is the temperature coefficient, is a negative sample, N is the number of negative samples. As Figure 3 shown, in the low-altitude illumination environment, the adaptive noise suppression mechanism effectively improves the signal-to-noise ratio and retains more details: the SNR is increased by 16.22 dB, and the detail retention rate is 24.80%.

[0079] As a preferred implementation manner, the specific steps of the said Step5 are:

[0080] Enhance the potential features of the image obtained after Step4.2 by means of the attention mechanism for the key regions (such as the edges and core parts of the target). During the training process, the model will automatically learn which regions are the most important for target recognition, so as to enhance the feature expression ability of these regions. The attention mechanism can be realized by the following formula:

[0081] ;

[0082] where, is the learned attention weight, is the feature representation of the i-th local region.

[0083] Furthermore, the specific steps of the said Step6 are:

[0084] Use the density-based clustering algorithm to group similar features into the same class, so as to identify and analyze new space targets. The density-based clustering algorithm can be realized by the following formula:

[0085] ;

[0086] where, is the j-th clustering center, is the sample set of the j-th clustering, is the sample point in the feature representation.

[0087] As a preferred implementation manner, the steps of using the non-regularized self-supervised learning mechanism to enhance the image of the target analyzed in Step6 in the said Step7 are:

[0088] Step7.1. Smooth the input low-illumination image by using the bilateral filtering technology to initially estimate the illumination map, which can be realized by the following formula:

[0089] ;

[0090] where, Indicates the preliminary illumination map, and S represents the input low-illumination image;

[0091] Step7.2. Further optimize the preliminary illumination map obtained just now using the self-supervised learning model to obtain a more accurate and consistent optimized illumination map, which can be achieved using the following formula:

[0092] ;

[0093] Among them, are model parameters obtained through optimized training; is the self-supervised learning model; is the preliminary illumination map, and S is the input low-illumination image;

[0094] Step7.3. Apply the correction factor r to adjust the illumination map to obtain the adjusted illumination map, which can be achieved using the following formula:

[0095] ;

[0096] Among them, r is the correction factor obtained by automatic learning during the training process, is the adjusted illumination map, is the optimized illumination map;

[0097] Step7.4. Calculate the reflectance map according to the Retinex theory, which can be achieved using the following formula:

[0098] ;

[0099] Among them, is the reflectance map, is the input low-illumination image, is the adjusted illumination map;

[0100] Take the reflectance map R as the enhanced image and visualize it by magnifying its intensity value, which can be achieved using the following formula:

[0101] ;

[0102] Among them, The function is used to magnify the intensity value of the reflectance map to make the resulting image brighter and clearer. As shown in Figure 4 , after the image enhancement step, the brightness and contrast of the image are significantly improved: the brightness is increased by 20.21%, the contrast is increased by 25.42%, and the processing time is 5.03 seconds.

[0103] In summary, the method for enhancing the contrast features of space targets with low illuminance based on self-supervised learning realizes the efficient enhancement and accurate detection of the contrast features of space targets in low-illuminance environments by introducing a self-supervised learning mechanism and an adaptive noise suppression technique. This method not only improves the efficiency and quality of image processing, enhances the robustness and accuracy of target detection, but also can meet the actual application requirements in various complex low-illuminance environments.

[0104] 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 its equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A method for enhancing contrast features of low - illumination space targets based on self - supervised learning, characterized in that It includes the following steps: Step 1. Collect spatial images in a low-illuminance space scene; Step 2. Perform data input and data augmentation operations on the low-illuminance spatial images collected in Step 1, including rotation and cropping; Step 3. Use a self-supervised contrast learning mechanism to extract features from the images after data augmentation in Step 2; Step 4. Use an adaptive noise suppression mechanism to perform dynamic noise suppression on the images after feature extraction in Step 3. The specific steps include: Step 4.

1. First, estimate the noise intensity in the local regions of the image, automatically identify the regions containing noise in the image using a self-supervised learning framework, and estimate and suppress the noise based on the noise distribution model of these regions. Secondly, introduce a noise weight map W to represent the noise intensity of each region. The noise suppression loss function is as shown in the formula: ； Among them, represents the feature of the i-th region in the input dim weak space image S, is the image after denoising feature of the corresponding region in, is the adaptive noise weight adjusted according to the regional noise intensity, is the indicator function; Step 4.

2. Further introduce a self-supervised contrast learning mechanism to perform feature learning and contrast optimization on the images after dynamic noise suppression. During the contrast learning process, design a mechanism including positive and negative sample contrast. The positive samples are generated through image enhancement, and the negative samples come from different images or background regions. The loss function of the contrast learning is as shown in the formula: ； wherein, and are the feature representations from the input image S and the enhanced image T respectively, is the temperature coefficient, is a negative sample, and N is the number of negative samples; Step 5. Use an attention mechanism to enhance the features of the key regions of the images after dynamic noise suppression in Step 4; Step 6. Use a density-based clustering algorithm to classify similar features of the images after feature enhancement in Step 5 and perform target recognition and analysis; Step 7. Use an unregularized self-supervised learning mechanism to perform image enhancement on the images after target analysis in Step 6. The specific steps are: Step 7.

1. Use bilateral filtering technology to smooth the input low-illuminance image and preliminarily estimate the illumination map, which is achieved through the following formula: ； Among them, represents the preliminary illumination map, and S represents the input low-illumination image; Step 7.

2. Use a self-supervised learning model to further optimize the preliminary illumination map just obtained, which is achieved through the following formula: ； Among them, are model parameters, obtained by optimizing training; is a self-supervised learning model; is a preliminary illumination map, and S is the input low-illuminance image; Step 7.

3. Apply a correction factor r to adjust the illumination map to obtain an adjusted illumination map, which is achieved through the following formula: ； where r is a correction factor, which is automatically learned through the training process. is the adjusted illumination map, and is the optimized illumination map; Step 7.

4. Calculate the reflectance map according to the Retinex theory, which is achieved through the following formula: ； Among them, is the reflectivity map, is the input low-illumination image, is the adjusted illumination map; Take the reflectance map R as the enhanced image and achieve visualization by amplifying its intensity value, which is achieved through the following formula: ； Among them, The function is used to amplify the intensity value of the reflectivity graph .

2. The method for enhancing the contrast features of low-illuminance space targets based on self-supervised learning according to claim 1, wherein The specific steps of Step 2 are: Step 2.

1. Perform geometric transformations on the faint spatial images, including rotation, translation, cropping, scaling, and flipping; Step 2.

2. Perform color space transformations on the faint spatial images after geometric transformations, including brightness adjustment, contrast adjustment, color perturbation, and grayscaling, and use these enhanced samples as "positive samples", while using samples from different images as "negative samples" for contrast.

3. The method for enhancing the contrast feature of low-illuminance space targets based on self-supervised learning according to claim 2, wherein The specific steps of Step 3 are: Take the enhanced data obtained after Step 2.2 as the "positive samples", and at the same time take samples from different images as the "negative samples" for comparison. Optimize the feature representation through the following formula for contrastive learning objectives: ; Among them, is the image feature representation extracted by the neural network, is the temperature parameter, is a negative sample, and N is the total number of negative samples.

4. The method for enhancing contrast features of low-illuminance space targets based on self-supervised learning according to claim 1, wherein The specific steps of Step 5 are: Enhance the key regions by means of the attention mechanism for the latent features of the images obtained after Step 4.

2. The attention mechanism is achieved through the following formula: ； Among them, is the learned attention weight, is the feature representation of the i-th local region.

5. The method for enhancing contrast features of low-illuminance space targets based on self-supervised learning according to claim 1, wherein The specific steps of Step 6 are: Use a density-based clustering algorithm to group similar features into the same class. The density-based clustering algorithm is implemented through the following formula: ； Among them, is the j-th clustering center, is the sample set of the j-th cluster, is the sample point in the feature representation.

Images