Infrared small target detection method based on robust unsupervised multi-feature representation

Abstract

The application provides an infrared small target detection method based on robust unsupervised multi-feature representation. The feature extraction accuracy is improved by designing a model based on robust unsupervised clustering, and the feature representation capability is improved by fusing pixel-level local contrast, distribution and direction gradient and other multi-features, so as to further improve the detection accuracy. The technical scheme is: 1) a three-layer filtering window structure is designed to quickly calculate the center weighted local contrast of each pixel; 2) threshold segmentation is used to extract candidate targets; 3) a model based on robust unsupervised clustering is constructed to extract target features from a complex background; 4) pixel-level multi-feature representation is performed on the candidate targets; 5) adaptive threshold segmentation is used to extract targets to obtain the final detection result. The method accurately mines the characteristic differences between the infrared small target and the background clutter, can effectively improve the target enhancement and background suppression capability, and further improves the performance of the infrared small target detection algorithm.

Description

Technical Field

[0001] This invention belongs to the field of image target detection technology, specifically relating to an infrared small target detection method based on robust unsupervised multi-feature representation. Background Technology

[0002] Infrared small target detection, a key technology in infrared search and track systems, plays a crucial role in numerous fields. Due to the need for long-range, early warning, small targets in infrared images typically occupy only a few pixels and lack texture detail, making practical infrared small target detection extremely difficult. Furthermore, infrared small targets are easily obscured by high-intensity backgrounds with sharp edges, and in applications such as UAV swarms, they are subject to interference from nearby targets, leading to reduced detection performance. Therefore, designing a robust, high-performance infrared small target detection method remains a valuable and challenging task.

[0003] Existing infrared small target detection methods can be broadly categorized into three types: The first type directly filters the image, such as maximum mean filtering (Max-Mean) and top-hat filtering. These methods are simple to implement but struggle to suppress complex backgrounds. The second type is based on component analysis, such as the Infrared Block Image Model (IPI) and the Non-convex Tensor Fiber Rank Approximation Model (NTFRA). These methods have good background suppression capabilities but struggle to handle sparse interference such as high-brightness noise points and have high computational complexity, thus limiting their detection performance. The third type is based on local feature representation, such as Local Contrast Measurement (LCM) and Adaptive Scale Image Patch-Based Local Contrast Measurement (ASPCM). These methods characterize the differences between the potential target and its surrounding background by designing local feature extraction models and feature representation methods, thereby achieving target detection. They exhibit good detection performance and interpretability, and have therefore been widely studied in recent years. To further improve detection performance, many researchers have focused on designing new feature extraction models, such as:

[0004] Patent document with application number (CN202110457402.4) proposes an infrared small target detection method and system based on adaptive scale image block weighted difference measurement. By designing a target scale estimation scheme based on edge detection, a feature extraction model that is adaptively adjusted according to the target scale is constructed, thereby improving the feature extraction accuracy and detection performance.

[0005] The patent document with application number (CN202110733585.8) proposes an infrared small target detection method based on pixel-level local contrast measurement. By presetting the number of seed points and cluster categories, the method uses a random walk algorithm to segment the candidate target and the surrounding background region, thereby achieving pixel-level subdivision of the target and background, thus extracting target features more accurately and improving detection accuracy.

[0006] The above detection methods based on local feature representation can improve the feature extraction accuracy to a certain extent, but still have the following shortcomings: (1) Existing feature extraction models usually assume that there is only a single target and a uniformly changing background in the local area. Therefore, when faced with complex background interference or other target interference, the robustness is insufficient, which makes the feature extraction accuracy decrease significantly; (2) Existing methods do not fully explore the characteristic differences between the target and the background, and cannot effectively represent the target and background features, making it difficult to further improve the detection performance. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention proposes an infrared small target detection method based on robust unsupervised multi-feature representation. On one hand, this invention designs a robust unsupervised clustering method to accurately segment the target and surrounding background interference, constructing a feature extraction model based on unsupervised clustering to improve feature extraction accuracy and robustness. On the other hand, this invention designs a multi-feature representation operator that fuses pixel-level local contrast, distribution, and directional gradient, effectively enhancing feature representation capabilities and improving detection accuracy. This invention overcomes the deficiencies of existing methods, effectively enhancing target detection while suppressing complex backgrounds, thus improving target detection performance.

[0008] To achieve the above objectives, the basic idea of ​​this invention is as follows: First, a three-layer filter window structure is designed to quickly calculate the center-weighted local contrast of each pixel; second, threshold segmentation is used to extract candidate targets; then, a robust unsupervised clustering model is constructed to accurately extract target features; next, pixel-level multi-feature representation is performed on the candidate targets; finally, adaptive threshold segmentation is used to extract the targets, and the final detection result is obtained.

[0009] The main objective of this invention is to address the problems of inaccurate feature extraction, insufficient feature representation capabilities, and limited detection accuracy in existing infrared small target detection methods. This invention improves feature extraction accuracy by designing a robust unsupervised clustering model and enhances feature representation capabilities by fusing multiple features such as pixel-level local contrast, distribution, and directional gradient, thereby further improving detection accuracy. This method accurately mines the characteristic differences between infrared small targets and background clutter, effectively improving target enhancement and background suppression capabilities, and further enhancing the performance of infrared small target detection algorithms.

[0010] The specific technical solution adopted by this invention to solve its technical problem is as follows:

[0011] An infrared small target detection method based on robust unsupervised multi-feature representation is characterized by employing a three-layer filter window structure to quickly calculate the center-weighted local contrast of each pixel; using threshold segmentation to extract candidate targets; then, constructing a model based on robust unsupervised clustering to accurately extract target features and perform pixel-level multi-feature representation on candidate targets; finally, using adaptive threshold segmentation to extract targets and obtain the final detection result.

[0012] Furthermore, the specific steps include:

[0013] Step 1: Input an infrared image I of size M×N, where the grayscale values ​​of the pixels in image I are represented as I(x,y), x=1,2,...,M,y=1,2,...,N;

[0014] Step 2: Construct a 3×3 pixel square target center layer TC, a 9×9 pixel square target attention layer TA, and a 15×15 pixel square background layer B, centered on each pixel (x, y) of the input image, from top to bottom and from left to right. Then, calculate the average gray value of each layer according to the following formula:

[0015]

[0016]

[0017]

[0018] Where N TC N TA and N B Let represent the number of pixels in regions TC, TA, and B, respectively; then calculate the center-weighted local contrast (CWLCM) of the current pixel (x, y) as follows:

[0019]

[0020] Once the CWLCM calculation for each pixel is complete, the CWLCM-enhanced image is obtained.

[0021] Step 3: Perform thresholding on the CWLCM-enhanced image to obtain a candidate target set:

[0022] Specifically, based on the enhanced image, the segmentation threshold is calculated:

[0023] Th cd =max{δ×M1,M k},

[0024] Among them, M1 and M kLet δ represent the maximum value and the k-th largest value in the image, respectively, where δ∈(0,1) and k∈[10,50] are empirical values; then, pixels with gray values ​​greater than the threshold are retained to obtain the candidate target pixel set S. CT ;

[0025] Step 4: Using set S one by one CT Centered on the pixel in the image, a circular unsupervised clustering region UCA with a diameter of 30 pixels is constructed. Then, the robust unsupervised clustering method based on DBSCAN is used for image segmentation to obtain the clustered local regions, including the central target region T, the surrounding background region Bi, and the surrounding target region Ci.

[0026] Step 5: Based on the clustering results in Step 4, perform pixel-level multi-feature representation on the candidate targets to enhance the targets and suppress complex backgrounds;

[0027] Step 6: Based on the multi-feature representation of each pixel measured in Step 5, construct an MFR enhanced image; then perform adaptive threshold segmentation on the MFR enhanced image to obtain the infrared small target detection results.

[0028] Furthermore, step 4 specifically includes the following sub-processes:

[0029] Step 4.1: Set the clustering parameters ε and MinPts.

[0030] Step 4.2: Construct a distance matrix for all pixels within the clustering region UCA. Where NS represents the number of candidate target pixels, and the value of the element (i, j) in the matrix is:

[0031]

[0032] Where θ i and I i Let represent the coordinate position and gray value of the i-th pixel in the input image corresponding to the pixel (x, y), d represent the Euclidean distance, and Th be the threshold function, defined as follows:

[0033]

[0034] Step 4.3: Calculate the number of pixels in the ε-neighborhood of each pixel using the following formula:

[0035] N ε (p i )={p j ∈D|d(θ i ,θ j )≤ε},

[0036] Where p i This represents the i-th pixel;

[0037] Step 4.4: Set conditions N to be met. ε (p i Pixels ≥ MinPts are labeled as core samples and numbered 1, 2, 3...N respectively. C The label represents the cluster category, where Nc represents the number of clusters after clustering;

[0038] Step 4.5: Find the samples that are density-connected to the core samples one by one, and label them as the categories corresponding to the core samples; for a core sample p i Its density-connected samples are defined as satisfying any one of three conditions: direct density reachability, density reachability, or density connectivity; where direct density reachability means that the sample is within the core sample p. i Within the ε-neighborhood; a density of up to indicates the existence of a relation chain p consisting of core objects. i ,p i+1 ,...,p i+n , satisfying p i Direct density can reach p i+1 Then p is called i+n It is p i Density-reachable samples; density-connected samples refer to two core samples p i and p j An intermediate sample p exists in the middle. n , satisfying p i+n Simultaneously, the density can reach sample p. i and p j ;

[0039] Step 4.6: Form the central target region T with pixels of the same category as the center pixel; set the clusters adjacent to the pixels in the target region as the surrounding background region Bi; set other clusters as the surrounding target region Ci; determine whether the following conditions are met in the current local region:

[0040] Condition 1: The number of pixels in region T is less than 81;

[0041] Condition 2: The total number of clusters is less than 9;

[0042] Condition 3: Parameter ε is less than 10.

[0043] If any condition is met, proceed to step 5; otherwise, adjust the parameter ε to half of its original value and continue executing steps 4.1 to 4.5.

[0044] Furthermore, step 5 specifically includes the following sub-steps:

[0045] Step 5.1: Express the local contrast features of the target according to the following formula:

[0046]

[0047] Where nb represents the number of surrounding background regions, m T and m Bi These represent the average intensities of regions T and Bi, respectively.

[0048] Step 5.2: Express the distribution characteristics of the target according to the following formula:

[0049]

[0050] Step 5.3: Divide the local region into eight directional regions with the included angle along the positive X-axis: 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°. Then, represent the target's directional gradient characteristics using the following formula:

[0051]

[0052] Among them MDG i The maximum gradient in the i-th directional region is calculated using the following formula:

[0053]

[0054] Where Di represents the i-th directional region, and DG is the directional gradient value, defined as:

[0055]

[0056] Where θ represents the direction angle of the region, and Δx and Δy represent the offsets, defined as:

[0057]

[0058] Where Round represents the floor function;

[0059] Step 5.4: Finally, use multiple features to represent the target characteristics:

[0060] MFR = MFR LCM ×MFR SF ×MFR DGF .

[0061] Furthermore, step 6 specifically includes the following sub-steps:

[0062] Step 6.1: Calculate the threshold T of the MFR-enhanced image. The calculation formula is as follows:

[0063] T = λ × Mx + (1 - λ) × Mn,

[0064] Where Mx and Mn represent the maximum and average gray values ​​of the MFR-enhanced image, respectively, and λ is a constant adjusted experimentally.

[0065] Step 6.2: Calculate the MFR enhancement image based on the threshold T to obtain the final detection result image R, where the grayscale values ​​of the pixels in image R are represented as R(x,y), x=1,2,...,M, y=1,2,...,N; the calculation formula is:

[0066]

[0067] The positions of non-zero pixels in the detection result image R are the positions of the infrared small targets.

[0068] Compared with the prior art, the present invention and its preferred embodiments have at least the following beneficial effects:

[0069] (1) Design a three-layer local filtering window structure, which can quickly and accurately measure local contrast, effectively extract candidate targets, and facilitate subsequent refined feature representation.

[0070] (2) A feature extraction model based on robust unsupervised clustering is proposed, which can intelligently segment targets and complex interferences, effectively improving the accuracy and robustness of feature extraction.

[0071] (3) A multi-feature representation method that integrates pixel-level local contrast, distribution and directional gradient is proposed, which can effectively enhance the target and suppress the background, thereby further improving the detection performance. Attached Figure Description

[0072] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0073] Figure 1 This is a general flowchart of an embodiment of the present invention;

[0074] Figure 2 This is a schematic diagram of the structure of the three-layer filter window in an embodiment of the present invention;

[0075] Figure 3 This is a schematic diagram of a feature extraction model based on unsupervised clustering in an embodiment of the present invention. Detailed Implementation

[0076] To make the features and advantages of this patent more apparent and understandable, specific embodiments are provided below for detailed explanation:

[0077] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0078] Reference Figure 1 The specific implementation of the present invention includes the following steps:

[0079] Step 1: Input an infrared image I of size M×N, where the grayscale values ​​of the pixels in image I are represented as I(x,y), x=1,2,...,M,y=1,2,...,N;

[0080] Step 2: Following the order from top to bottom and from left to right, construct the following structures centered on the input image pixels (x, y): Figure 2 The target center layer (denoted by TC), a target attention layer (denoted by TA), and a background layer (denoted by B), which are 3×3 pixel squares, are shown in the diagram. Then, the average gray value of each layer is calculated according to the following formula:

[0081]

[0082]

[0083]

[0084] Where N TC N TA and N B These represent the number of pixels in regions TC, TA, and B, respectively. Then, the center-weighted local contrast ratio (CWLCM) of the current pixel (x, y) is calculated as follows:

[0085]

[0086] Once the CWLCM calculation for each pixel is complete, the CWLCM-enhanced image can be obtained.

[0087] Step 3: Perform thresholding segmentation on the CWLCM-enhanced image to obtain a candidate target set. Specifically, calculate the segmentation threshold based on the enhanced image:

[0088] Th cd =max{δ×M1,M k},

[0089] Among them, M1 and M k Let δ represent the maximum value and the k-th largest value in the image, respectively, where δ∈(0,1) and k∈[10,50] are manually set empirical values. Then, pixels with grayscale values ​​greater than the threshold are retained to obtain the candidate target pixel set S. CT .

[0090] Step 4: Using set S one by one CT Centered on the pixels in the image, a circular unsupervised clustering region (UCA) with a diameter of 30 pixels is constructed. Then, the image is segmented using the unsupervised clustering method DBSCAN, resulting in the following: Figure 3The local region shown includes the central target region T, the surrounding background region Bi, and the surrounding target region Ci. Specifically, it includes the following sub-processes:

[0091] Step 4.1: Set the clustering parameters ε and MinPts.

[0092] Step 4.2: Construct a distance matrix for all pixels within the clustering region UCA. Where NS represents the number of candidate target pixels, and the value of the element (i, j) in the matrix is:

[0093]

[0094] Where θ i and I i Let represent the coordinate position and gray value of the i-th pixel in the input image corresponding to the pixel (x, y), d represent the Euclidean distance, and Th be the threshold function, defined as follows:

[0095]

[0096] Step 4.3: Calculate the number of pixels in the ε-neighborhood of each pixel using the following formula:

[0097] N ε (p i )={p j ∈D|d(θ i ,θ j )≤ε},

[0098] Where p i This represents the i-th pixel.

[0099] Step 4.4: Set conditions N to be met. ε (p i Pixels ≥ MinPts are labeled as core samples and numbered 1, 2, 3...N respectively. C The label represents the cluster category, where Nc represents the number of clusters.

[0100] Step 4.5: Find the samples that are density-connected to the core samples one by one, and label them as the categories corresponding to the core samples. For a core sample p... i Its density-connected samples are defined as satisfying any one of three conditions: direct density reachability, density reachability, or density connectivity. Direct density reachability indicates that a sample is within the core sample p. i Within the ε-neighborhood; a density of up to indicates the existence of a relation chain p consisting of core objects. i ,p i+1 ,...,p i+n , satisfying p i Direct density can reach pi+1 Then p is called i+n It is p i Density-reachable samples; density-connected samples refer to two core samples p i and p j An intermediate sample p exists in the middle. n , satisfying p i+n Simultaneously, the density can reach sample p. i and p j .

[0101] Step 4.6: Form the central target region T with pixels of the same category as the center pixel of the circle, set the clusters adjacent to the pixels of the target region as the surrounding background region Bi, and set the other clusters as the surrounding target region Ci.

[0102] Step 5: Based on the clustering results in Step 4, perform multi-feature representation on the candidate targets to enhance the targets and suppress complex backgrounds. This includes the following sub-steps:

[0103] Step 5.1: Express the local contrast features of the target according to the following formula:

[0104]

[0105] Where nb represents the number of surrounding background regions, m T and m Bi These represent the average intensity of regions T and Bi, respectively.

[0106] Step 5.2: Express the distribution characteristics of the target according to the following formula:

[0107]

[0108] Step 5.3: Divide the local region into eight directional regions with the included angle along the positive X-axis: 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°. Then, represent the target's directional gradient characteristics using the following formula:

[0109]

[0110] Among them MDG i The maximum gradient in the i-th directional region can be calculated using the following formula:

[0111]

[0112] Where D i Let DG represent the region in the i-th direction, and DG be the directional gradient value, defined as:

[0113]

[0114] Where θ represents the direction angle of the region, and Δx and Δy represent the offsets, defined as:

[0115]

[0116] Where Round represents the floor function.

[0117] Step 5.4: Finally, use multiple features to represent the target characteristics:

[0118] MFR = MFR LCM ×MFR SF ×MFR DGF .

[0119] Step 6: Based on the multi-feature representations of each pixel measured above, construct an MFR enhanced image. Perform adaptive threshold segmentation on the MFR enhanced image to obtain the infrared small target detection results, specifically including the following sub-steps:

[0120] Step 6.1: Calculate the threshold T of the MFR-enhanced image. The calculation formula is as follows:

[0121] T = λ × Mx + (1 - λ) × Mn, where Mx and Mn represent the maximum and average gray values ​​of the MFR-enhanced image, respectively, and λ is a constant that can be adjusted experimentally, with a recommended setting of 0.2.

[0122] Step 6.2: Calculate the MFR enhancement image based on the threshold T mentioned above to obtain the final detection result image R, where the grayscale values ​​of the pixels in image R are represented as R(x,y), x=1,2,...,M, y=1,2,...,N. The calculation formula is:

[0123]

[0124] The positions of non-zero pixels in the detection result image R are the positions of the infrared small targets.

[0125] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0129] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

[0130] This patent is not limited to the above-described preferred embodiments. Anyone can derive other forms of infrared small target detection methods based on robust unsupervised multi-feature representation under the guidance of this patent. All equivalent changes and modifications made within the scope of this patent application shall fall within the scope of this patent.

Claims

1. A robust unsupervised multi-feature representation-based infrared small target detection method, characterized by: Includes the following steps: Step 1: Input an infrared image of size M×N , where the image The grayscale value of a pixel is represented as ; Step 2: Construct a 3×3 pixel square target center layer TC, a 9×9 pixel square target attention layer TA, and a 15×15 pixel square background layer B, centered on each pixel (x, y) of the input image, from top to bottom and from left to right. Then, calculate the average gray value of each layer according to the following formula: in , and Let represent the number of pixels in regions TC, TA, and B, respectively; then calculate the center-weighted local contrast (CWLCM) of the current pixel (x, y) as follows: Once the CWLCM calculation for each pixel is complete, the CWLCM-enhanced image is obtained. Step 3: Perform thresholding on the CWLCM-enhanced image to obtain a candidate target set: Specifically, based on the enhanced image, the segmentation threshold is calculated: Among them, M1 and M k These represent the maximum value and the k-th largest value in the image, respectively. and The threshold is set as an empirical value; then pixels with grayscale values ​​greater than the threshold are retained to obtain a set of candidate target pixels. ; Step 4: One by one with sets Centered on the pixel in the image, a circular unsupervised clustering region UCA with a diameter of 30 pixels is constructed. Then, the robust unsupervised clustering method based on DBSCAN is used for image segmentation to obtain the clustered local regions, including the central target region T, the surrounding background region Bi, and the surrounding target region Ci. Step 5: Based on the clustering results in Step 4, perform pixel-level multi-feature representation on the candidate targets to enhance the targets and suppress complex backgrounds; Step 6: Based on the multi-feature representation of each pixel measured in Step 5, construct an MFR enhanced image; then perform adaptive threshold segmentation on the MFR enhanced image to obtain the infrared small target detection results; Step 5 specifically includes the following sub-steps: Step 5.1: Express the local contrast features of the target according to the following formula: Where nb represents the number of surrounding background regions, m T and m Bi These represent the average intensities of regions T and Bi, respectively. Step 5.2: Express the distribution characteristics of the target according to the following formula: Step 5.3: Divide the local region into eight directional regions with the included angle along the positive X-axis: 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°. Then, represent the target's directional gradient characteristics using the following formula: in The maximum gradient in the i-th directional region is calculated using the following formula: Where D i Let DG represent the ith directional region, where DG is the directional gradient value, defined as: in Indicates the direction and angle of the area. and The offset is represented and defined as: Where Round represents the floor function; Step 5.4: Finally, use multiple features to represent the target characteristics: 。 2. The infrared small target detection method based on robust unsupervised multi-feature representation according to claim 1, characterized in that: Step 4 specifically includes the following sub-processes: Step 4.1: Set clustering parameters and , Step 4.2: Construct a distance matrix for all pixels within the clustering region UCA. Where NS represents the number of candidate target pixels, and the value of the element (i, j) in the matrix is: in and Let represent the coordinates (x, y) and grayscale value of the i-th pixel in the input image, d represent the Euclidean distance, and Th be the threshold function, defined as follows: Step 4.3: Calculate the number of pixels in the ε-neighborhood of each pixel using the following formula: in This represents the i-th pixel; Step 4.4: [The following appears to be a separate, unrelated section:] ...and meet the conditions The pixels are labeled as core samples and numbered 1, 2, 3...N respectively. C The label represents the cluster category, where Nc represents the number of clusters after clustering; Step 4.5: Find the samples that are density-connected to the core samples one by one, and label them as the categories corresponding to the core samples; for a core sample Its density-connected samples are defined as those satisfying any one of three conditions: direct density reachability, density reachability, or density connectivity; where direct density reachability indicates that the sample is within the core sample. Within the ε-neighborhood; a density that can be reached indicates the existence of a relational chain composed of core objects. ,satisfy Direct density can reach Then it is called yes Density-reachable samples; density-connected samples refer to two core samples. and An intermediate sample exists in ,satisfy Simultaneously, the density can reach the sample and ; Step 4.6: Form the central target region T with pixels of the same category as the center pixel; set the clusters adjacent to the pixels in the target region as the surrounding background region Bi; set other clusters as the surrounding target region Ci; determine whether the following conditions are met in the current local region: Condition 1: The number of pixels in region T is less than 81; Condition 2: The total number of clusters is less than 9; Condition 3: Parameters Less than 10, If any condition is met, proceed to step 5; otherwise, set the parameters... Adjust it to half of the original value and continue with steps 4.1 to 4.

5.

3. The infrared small target detection method based on robust unsupervised multi-feature representation according to claim 2, characterized in that: Step 6 specifically includes the following sub-steps: Step 6.1: Calculate the threshold for the MFR-enhanced image. The calculation formula is as follows: ， in and These represent the maximum and average gray values ​​of the MFR-enhanced image, respectively. It is a constant adjusted experimentally; Step 6.2: Based on the threshold The MFR-enhanced image is calculated to obtain the final detection result image. , where the image The grayscale value of a pixel is represented as The calculation formula is: Detection result image The position of the non-zero pixel is the position of the infrared small target.

Images