Infrared Small Target Detection Method Based on Gradient Direction Difference and Thermonuclear Diffusion

Abstract

This invention discloses a method for detecting small infrared targets under complex background conditions, comprising the following steps: calculating the local gradient direction of the target image using a window method; weighting the residual saliency image based on the local gradient direction of the target image; extracting segmented regions from the weighted residual saliency image based on brightness information and defining potential target seed points based on local features; enhancing the energy response of suspicious regions through thermal kernel diffusion, thereby improving the contrast between the target and the background; describing pixels considered to belong to the target region in the segmented thermal kernel diffusion energy map, amplifying the difference between the target region and the background region, and generating an enhancement map; fusing the weighted residual saliency image and the enhancement map, and achieving the final target saliency map through adaptive segmentation. This invention has the advantages of strong robustness, low false alarm rate, and high detection accuracy.

Description

Technical Field

[0001] This invention belongs to the field of small target detection technology, specifically an infrared small target detection method based on gradient direction difference and thermonuclear diffusion. Background Technology

[0002] Infrared imaging systems are widely used in target detection and identification tasks because they can operate in low light and harsh environments. However, targets in infrared images are typically small in size, weak in brightness, and have limited structural information. They are also easily obscured by complex backgrounds and strong noise interference, significantly increasing the difficulty of detection.

[0003] Existing infrared small target detection methods mainly include filtering-based methods, low-rank and sparse decomposition-based methods, and saliency analysis-based methods. Filtering methods are effective in suppressing backgrounds, but they are prone to false detections in scenes with strong edges or textures. Low-rank and sparse decomposition methods can separate targets from the background to some extent, but their computational complexity is high, and they lack robustness in the face of strong noise interference. Saliency analysis-based methods rely on local contrast features and are difficult to effectively distinguish between targets and backgrounds with similar gradient distributions.

[0004] Therefore, in order to address the problems of insufficient detection accuracy, high false alarm rate, and blurred target boundaries of infrared small targets under complex backgrounds and strong noise conditions, there is an urgent need for a detection method that can make full use of the statistical features of gradient direction and combine regional diffusion characteristics to achieve accurate segmentation and enhanced detection of infrared small targets. Summary of the Invention

[0005] The purpose of this invention is to overcome the problems of insufficient detection accuracy, high false alarm rate and blurred target boundaries in existing infrared small target detection methods under complex background conditions. This invention provides an infrared small target detection method based on gradient direction difference and thermonuclear diffusion. This method can effectively distinguish the statistical differences between small targets and background structural regions, enhance target saliency, and suppress interference from complex backgrounds, thereby achieving high-precision infrared small target detection with a low false alarm rate.

[0006] The technical solution to achieve the objective of this invention is as follows: an infrared small target detection method based on gradient direction difference and thermonuclear diffusion, comprising the following steps:

[0007] The local gradient direction of the target image is calculated using the window method;

[0008] The residual significance image is weighted based on the local gradient direction of the target image;

[0009] In the weighted residual saliency image, segmentation regions are extracted based on brightness information, and potential target seed points are defined based on local features;

[0010] Enhance the energy response of suspicious areas through thermonuclear diffusion to distinguish targets from the background;

[0011] By constructing new boundary strength, anisotropic cumulative distance, and compactness descriptions, the pixels considered to belong to the target region in the segmented thermonuclear diffusion energy map are described, the difference between the target region and the background region is amplified, and an enhanced map is generated.

[0012] The weighted residual saliency image is fused with the enhancement map, and the final target saliency map is obtained through an adaptive segmentation operation.

[0013] Compared with the prior art, the present invention has the following advantages:

[0014] 1) This invention effectively distinguishes the structural features of small targets from complex backgrounds by introducing gradient direction statistical difference analysis, thereby improving the accuracy of detection;

[0015] 2) This invention utilizes a thermonuclear diffusion model to enhance the energy response of suspicious regions, thereby improving the contrast between the target and the background;

[0016] 3) This invention further improves the separability of the target and the background through an enhancement mechanism that combines boundary strength, isotropic cumulative distance and compactness description;

[0017] 4) The invention has strong overall robustness, good suppression ability against noise and structural interference, low false alarm rate, and high detection accuracy;

[0018] 5) The algorithm structure of this invention is clear, the computational efficiency is high, and it is easy to implement in engineering and integrate into systems. Attached Figure Description

[0019] Figure 1 This is the main flowchart of the infrared small target detection method based on gradient direction difference and thermonuclear diffusion of the present invention.

[0020] Figure 2 A schematic diagram of the directional gradient statistics for the target and background.

[0021] Figure 3 This is a schematic diagram illustrating the selection of seed points.

[0022] Figure 4 This is a diagram showing the segmentation of thermonuclear diffusion.

[0023] Figure 5 These are two cases with different levels of compactness. Detailed Implementation

[0024] like Figure 1 As shown, an infrared small target detection method based on gradient direction difference and thermonuclear diffusion includes the following steps:

[0025] 1) The local gradient direction of the target image is calculated using the window method. The specific method is as follows:

[0026] The local gradient direction of the target image is calculated using the window method. Specifically, in , , , , , , and Eight gradient directions are defined within a certain range. The local gradient directions of the target image are calculated. :

[0027]

[0028] in, This represents the number of pixels in the i-th direction. This represents the average number of pixels in each direction. Local gradient directions (LGD) are calculated across the entire image using a sliding window. These defined gradient directions can assess the uniformity of gradient directions in local regions.

[0029] like Figure 2 As shown, local blocks containing small targets exhibit anisotropic uniformity in the regional gradient direction. However, for high-intensity structural edges and cluttered edge regions, the gradient direction tends to converge in a specific direction.

[0030] 2) Based on the local gradient direction of the target image, the residual significance image is weighted. The specific method is as follows:

[0031] The significance image of the residuals is calculated using the following formula. :

[0032]

[0033]

[0034]

[0035]

[0036]

[0037] In the formula, Indicates Fourier transform; Indicates the inverse Fourier transform; It is a frequency domain smoothing kernel; "*" indicates a convolution operation; For the extracted target image The amplitude spectrum; For the extracted target image The phase spectrum; The transformed log spectrum; The difference between the log spectrum and its mean-filtered value is the spectral residual; the spectral residual... and phase The significance image of the residuals can be obtained by performing a two-dimensional inverse Fourier transform. .

[0038] Utilizing the local gradient direction of the target image The image shows the significance of the residuals. Weighting:

[0039]

[0040]

[0041] In the formula, the symbol Represents the Adama product. This represents the significance image of the weighted residuals.

[0042] 3) Extract segmentation regions from the weighted residual saliency image based on brightness information, and define potential target seed points based on local features. The specific method is as follows:

[0043] Figure 3 This is a schematic diagram illustrating the definition of seed points. First, an adaptive thresholding method is used to segment the weighted residual saliency image of the region. The adaptive threshold for segmentation is defined. for:

[0044]

[0045] in and Significance image of the weighted residuals The mean and standard deviation, The empirical value was set to 3 in the experiment. The specific operation for each pixel in the image is as follows:

[0046] Calculate the binary segmentation map Uniformity of distances in four directions for each connected region :

[0047]

[0048]

[0049] in, Corresponding to the i-th direction, from the pixel coordinates (x, y) of the center point of the segmented region to the pixel coordinates of the farthest boundary point in that direction. distance, This represents the average distance in four directions. Indicates the distance from the center point of the segmented region in the i-th direction. The farthest coordinate, if calculated using distances from the four directions, shows uniformity. Smaller than the set value Then the coordinates of the center point of the segmented region are considered to be To obtain the coordinates of the initial suspected target seed points, a preliminary seed point map is generated. ,Will The corresponding preliminary suspected target seed point coordinates are set to 1:

[0050]

[0051] Furthermore, to further constrain the target shape, the maximum value among the maximum distances in the four directions is selected. Defined as:

[0052]

[0053] like Pixels are considered to be the center point of the segmented region. The target scale constraint is met, resulting in the second stage of screening. :

[0054]

[0055] Finally, the results of the two screening stages are multiplied point by point to obtain the final potential target seed point map. :

[0056]

[0057] 4) Thermonuclear energy diffusion:

[0058] For potential target seed points, using their center positions as a reference, corresponding local image patches are extracted from the original infrared image. The size of each local image patch is [size missing]. Construct a degree matrix between each pixel and adjacency matrix , where the degree matrix elements in Specifically:

[0059]

[0060] in, , These represent the first and second parts of the image patch, respectively. and the The intensity of a pixel.

[0061] Adjacency matrix elements in Specifically:

[0062]

[0063] in and Indicates the position of two pixels, parameters This is a distance threshold that controls whether two pixels are connected in the graph. The set parameters control the similarity of pixels under different contrast conditions.

[0064] The adjacency matrix is ​​constructed based on the local image patch where the potential target seed point is located. Sum-degree matrix The Graph Laplace matrix can be calculated. And carry out thermonuclear energy diffusion:

[0065]

[0066] in This represents the heat core, describing the change in energy diffusion over time within the local image patch where the potential target seed point is located. Representing time, thus obtaining the thermonuclear diffusion energy diagram. .

[0067] Energy diagram of thermonuclear diffusion A threshold was set. The threshold, used to distinguish between the target and the background, is defined as follows:

[0068]

[0069] in The maximum energy value representing the thermonuclear diffusion energy map. The minimum energy value representing the thermonuclear diffusion energy map is used. Pixels that are higher than or equal to the threshold are considered to belong to the target region, while pixels that are lower than the threshold are considered to belong to the background region.

[0070] 5) Target region feature description and enhancement:

[0071] a) Boundary strength description

[0072] The boundary strength is defined by the following formula:

[0073] In the formula, and These represent the average pixel energy of pixels belonging to the target region and the average pixel energy of pixels belonging to the background region, respectively.

[0074] With appropriate boundary weights, significant differences between the inside and outside of a strong boundary are typically expressed as orders of magnitude differences. For example, an energy difference of 10^3 or even 10^5 can effectively characterize a strong boundary. To bridge this descriptive gap, logarithmic functions are used. For instance, energy differences of 10^3 and 10^5 correspond to logarithmic values ​​of 6.9 and 11.5, respectively. In cases of ambiguous boundaries, the average energy ratio between the inside and outside approaches 1, or approaches 0 when described using logarithmic functions.

[0075] b) Regional compactness

[0076] When segmenting local regions using the hot core diffusion model, pixels extracted from small targets tend to exhibit a compact and clustered state. Specifically, each pixel has a nearest neighbor pixel with a distance of 1. On the other hand, cluttered edge regions or jagged regions may lead to pixel dispersion. To evaluate the compactness of a region, a distance matrix is ​​constructed. It captures the distance between pixels within the region:

[0077] in, This represents the distance between the i-th and j-th pixels in the segmented thermonuclear diffusion energy map, which are considered to belong to the target region. This indicates the number of pixels belonging to the target region.

[0078] Construct an adjacency matrix based on the distance matrix. :

[0079]

[0080] Subsequently, by performing connectivity analysis on the undirected graph, starting from the unvisited node, each traversal of the undirected graph constructed by the adjacency matrix reaches all nodes that constitute a connected component, and the number of traversals is the number of connected components. .

[0081] Define compactness :

[0082]

[0083] Figure 5 The study demonstrates compactness in two different scenarios, revealing that small target areas typically exhibit better compactness and are clearly distinguishable from background clutter.

[0084] 3) Cumulative distance around isotropic regions

[0085] The thermonuclear diffusion energy map is divided into several directional regions based on the angle of each pixel relative to the center point of the map. The span of each directional region is... ;

[0086] Calculate the cumulative distance of each pixel within each directional region that is considered to belong to the target region.

[0087]

[0088] in, Indicates the first The cumulative distance of all pixels within a given directional region that are considered to belong to the target region; Indicates the first All pixels within a given area that are considered to belong to the target area. This represents the coordinates of the center point in the thermonuclear diffusion energy diagram. This indicates the first region within that direction that is considered to belong to the target region. The coordinates of one pixel;

[0089] The isotropic uniformity of small targets is calculated by accumulating distances:

[0090]

[0091] in, This represents the average cumulative distance in eight directions, where, The scale normalization constant (threshold z-value, used to normalize the scale) Mapped to Interval; when hour, ;when hour, .

[0092] By fusing boundary strength, compactness, and various uniformities, an enhanced image is obtained. :

[0093] 6) Fusion of the weighted residual significance image and the enhancement image:

[0094] Bundle and By combining them, we obtain the final image. .

[0095] An adaptive threshold segmentation is performed on the weighted augmentation map to obtain the final target saliency map. We define the adaptive threshold for segmentation. for:

[0096]

[0097] in and For weighted augmented graph The mean and standard deviation, The empirical value was set to 3 in the experiment. Final target saliency map. The settings are as follows:

[0098] .

Claims

1. A method for detecting small infrared targets based on gradient direction differences and thermonuclear diffusion, characterized in that, Includes the following steps: The local gradient direction of the target image is calculated using the window method; The residual significance image is weighted based on the local gradient direction of the target image; In the weighted residual saliency image, segmentation regions are extracted based on brightness information, and potential target seed points are defined based on local features; Enhance the energy response of suspicious areas through thermonuclear diffusion to distinguish targets from the background; By constructing new boundary strength, anisotropic cumulative distance, and compactness descriptions, the pixels considered to belong to the target region in the segmented thermonuclear diffusion energy map are described, the difference between the target region and the background region is amplified, and an enhanced map is generated. The weighted residual saliency image is fused with the enhancement map, and the final target saliency map is obtained through an adaptive segmentation operation.

2. The infrared small target detection method based on gradient direction difference and thermonuclear diffusion according to claim 1, characterized in that, The specific method for calculating the local gradient direction of a target image using the window method is as follows: Define eight gradient directions for the target image, and calculate the local gradient directions of the target image. : in, Indicates the first Number of pixels in each direction, This represents the average number of pixels in each direction.

3. The infrared small target detection method based on gradient direction difference and thermonuclear diffusion according to claim 1, characterized in that, The specific method for weighting the residual significance image based on the local gradient direction of the target image is as follows: First, analyze the residual significance image. Perform the calculation: In the formula, Indicates Fourier transform; Indicates the inverse Fourier transform; It is a frequency domain smoothing kernel; "*" indicates a convolution operation; For the extracted target image The amplitude spectrum; For the extracted target image The phase spectrum; The transformed log spectrum; For spectral residuals; This is a residual significance graph; Utilizing the local gradient direction of the target image For residual significance images Weighting: In the formula, the symbol Represents the Adama product. This represents the significance image of the weighted residuals.

4. The infrared small target detection method based on gradient direction difference and thermonuclear diffusion according to claim 1, characterized in that, The specific method for performing binary segmentation based on brightness information in the weighted residual saliency image and defining potential target seed points based on local features is as follows: An adaptive thresholding method is used to perform binary segmentation on the weighted residual significance image. The specific operation for each pixel in the image is as follows: In the formula, Image representing the significance of the weighted residuals In pixel coordinates Pixel value at; Calculate the binary segmentation map Uniformity of distances in four directions for each connected region : in, Corresponding to the i-th direction, from the pixel coordinates (x, y) of the center point of the segmented region to the pixel coordinates of the farthest boundary point in that direction. distance, This represents the average distance in four directions. Indicates the distance from the center point of the segmented region in the i-th direction. The farthest coordinate, if calculated using distances from the four directions, shows uniformity. Smaller than the set value Then the coordinates of the center point of the segmented region are considered to be To obtain the coordinates of the initial suspected target seed points, a preliminary seed point map is generated. ,Will The corresponding preliminary suspected target seed point coordinates are set to 1: Take the maximum value among the distances in the four directions. Defined as: like Pixels are considered to be the center point of the segmented region. The target scale constraint is met, resulting in the second stage of screening. : The results of the two screening stages are multiplied point by point to obtain the final potential target seed point map. : 。 5. The infrared small target detection method based on gradient direction difference and thermonuclear diffusion according to claim 4, characterized in that, Adaptive threshold for segmentation Specifically: in and Significance image of the weighted residuals The mean and standard deviation, The set empirical value.

6. The infrared small target detection method based on gradient direction difference and thermonuclear diffusion according to claim 1, characterized in that, The specific method for distinguishing targets from background by enhancing the energy response of suspicious regions through thermonuclear diffusion is as follows: For potential target seed points, using their center positions as a reference, corresponding local image patches are extracted from the original infrared image. The size of each local image patch is [size missing]. Construct a degree matrix between each pixel and adjacency matrix , where the degree matrix elements in Specifically: in, , These represent the first and second parts of the image patch, respectively. and the Pixel intensity; The adjacency matrix is ​​constructed based on the local image patch where the potential target seed point is located. Sum-degree matrix The Graph Laplacian matrix is ​​calculated. Thermonuclear energy diffusion is then performed on local image patches to obtain a thermonuclear diffusion energy map. ; Energy diagram of thermonuclear diffusion Set a threshold Used to distinguish between the target and the background, pixels that are higher than or equal to a threshold are considered to belong to the target area, while pixels that are lower than the threshold are considered to belong to the background area.

7. The infrared small target detection method based on gradient direction difference and thermonuclear diffusion according to claim 6, characterized in that, Adjacency matrix elements in Specifically: in and Indicates the position of two pixels. Distance threshold The parameters are set.

8. The infrared small target detection method based on gradient direction difference and thermonuclear diffusion according to claim 1, characterized in that, By constructing new descriptions of boundary strength, compactness, and uniformity, pixels considered to belong to the target region in the segmented thermonuclear diffusion energy map are described, amplifying the difference between the target region and the background region. The specific method for generating the enhanced map is as follows: The boundary strength is defined by the following formula: In the formula, and These represent the average pixel energy of pixels belonging to the target region and the average pixel energy of pixels belonging to the background region, respectively. Then, construct the distance matrix. : in, This represents the distance between the i-th and j-th pixels in the segmented thermonuclear diffusion energy map, which are considered to belong to the target region. This indicates the number of pixels belonging to the target region; Pixels belonging to the target region in the segmented thermonuclear diffusion energy map are considered nodes in the map. At this time, a connection is established between two pixels, thereby constructing an adjacency matrix. : Subsequently, by performing connectivity analysis on the undirected graph, starting from the unvisited node, each traversal of the undirected graph constructed by the adjacency matrix reaches all nodes that constitute a connected component, and the number of traversals is the number of connected components. . Define compactness : The thermonuclear diffusion energy map is divided into several directional regions based on the angle of each pixel relative to the center point of the map. The span of each directional region is... ; Calculate the cumulative distance of each pixel within each directional region that is considered to belong to the target region: in, Indicates the first The cumulative distance of all pixels within a given directional region that are considered to belong to the target region; Indicates the first All pixels within a given directional region that are considered to belong to the target region. This represents the coordinates of the center point in the thermonuclear diffusion energy diagram. Indicates the first Within the area considered to belong to the target area, the [number]th [location]... The coordinates of one pixel; The isotropic uniformity of small targets is calculated by accumulating distances. : in, This represents the average cumulative distance in eight directions. The scale normalization constant threshold is used to... Mapped to Interval; when hour, ;when hour, ; By fusing boundary strength, compactness, and isotropic homogeneity, an enhanced map is obtained. : 。 9. The infrared small target detection method based on gradient direction difference and thermonuclear diffusion according to claim 1, characterized in that, The specific method for fusing the weighted residual saliency image with the enhancement image and obtaining the final target saliency map through adaptive segmentation is as follows: Enhanced image Significance of weighted residuals Combine to obtain a weighted enhanced graph. ; Adaptive thresholding is applied to the weighted augmentation map to obtain the final target saliency map. Specifically: In the formula, For the adaptive threshold of segmentation, For weighted augmented graph The pixel value at pixel coordinates (x, y). For the final target saliency map The pixel value at pixel coordinates (x, y).

10. The infrared small target detection method based on gradient direction difference and thermonuclear diffusion according to claim 9, characterized in that, Adaptive threshold for segmentation for: in and For weighted augmented graph The mean and standard deviation, The set empirical value.

Images