Infrared small target detection method based on expansion contrast and structure tensor

Abstract

The invention provides an infrared small target detection method based on expansion contrast and structure tensor. Because the infrared weak target occupies few pixels, the signal-to-noise ratio of the target is low, the target is easy to be interfered by clutters and is extremely easy to submerge, and the detection difficulty is high. The method comprises the following steps: firstly, designing an expansion template, and quickly calculating an expansion contrast ratio; secondly, designing target representation by using image local structure tensor prior, constructing a weight function, amplifying the difference of a target in a local change rate, and enhancing the target; and finally, weighting the two to obtain a final saliency map. The method aims to solve the problem of infrared weak target detection of an infrared search and tracking system in a complex scene, and is suitable for the related fields of infrared security and protection and the like.

Description

Technical Field

[0001] The invention relates to the technical field of image processing and proposes an infrared small target detection method based on expansion contrast and structure tensor. Background Art

[0002] The infrared imaging system works in the infrared band and has the characteristics of being passive and non-contact, and not restricted by day or night. It can be applied to infrared security and other fields. Infrared weak targets are one of the key technologies of infrared search and tracking systems. Due to long-distance imaging, the target image size is very small, lacks shape and texture information, and the target radiation intensity is weak. When there are a large number of clutter and noise interferences with similar target characteristics in the scene, it is very difficult to extract the effective features of the target from the complex and blurred background. In addition, it is very likely that a large number of false targets with similar characteristics to the real target will coexist. In summary, infrared weak target detection under complex backgrounds is still a major challenge for current infrared search and tracking systems.

[0003] In recent years, infrared weak target detection methods have emerged continuously. Based on the background consistency method, various filters are designed according to prior information to estimate the background. Therefore, the choice of filter will directly affect the detection performance. It maintains effective detection performance under smooth background and has low complexity. However, when the scene is complex, this type of method has a high false alarm rate. The method based on low-rank decomposition uses the non-local autocorrelation of the background to transform the traditional infrared target detection problem into an optimization problem with a low-rank matrix and a sparse matrix. These methods converge through global optimization, which takes a long time and relies heavily on the designer to provide a large number of hyperparameters. The false alarm rate is high in complex scenes, and the false alarm rate of dark targets is also high. The HVS-based method manually designs feature descriptors, uses the difference between the local area of ​​the target and the background to construct local contrast, enhance the target and detect it. The advantages of this type of method are simple principles and more flexible representation of weak target features. However, there are many interferences in complex backgrounds. Among them, target similarity interference will also be enhanced, making it difficult to suppress false alarms. In addition, when the target signal-to-noise ratio is low, the difference is difficult to make the target stand out, and the detection performance is constrained and still needs to be improved.

[0004] In summary, various methods have designed a variety of feature descriptors to efficiently characterize the target or the difference between the target and the background, thereby enhancing the robustness of detection performance in complex scenes. Summary of the invention

[0005] The purpose of the present invention is to provide an infrared dim small target detection method based on expansion contrast and structure tensor, which is used to solve the problem of low target signal-to-noise ratio and high detection false alarm rate. The method uses the target feature fusion of the expansion contrast weighted structure tensor feature function. The method of the present invention is aimed at the infrared dim small target detection problem of infrared search and tracking system in complex scenes, and is particularly suitable for related fields such as infrared security.

[0006] The technical solution of the present invention is as follows:

[0007] An infrared small target detection method based on expansion contrast and structure tensor, characterized in that it comprises the following steps:

[0008] First, a dilation template is designed to quickly calculate the dilation contrast;

[0009] Secondly, the local structure tensor prior of the image is used to design the target representation and construct the weight function to amplify the difference in the local change rate of the target and enhance the target;

[0010] Finally, the two are weighted to obtain the final saliency map.

[0011] The infrared small target detection method based on expansion contrast and structure tensor is characterized by comprising the following steps:

[0012] A. Dilation operation of the original image:

[0013]

[0014] Among them, I raw is the original image, ΔB is the structural element, is the expansion operation; I dilation is the image after the dilation operation, and then the difference joint contrast is constructed with the original image to obtain the significant feature map:

[0015]

[0016] DCM is the dilated contrast saliency map;

[0017] B. Calculate the local structure tensor of the original image

[0018]

[0019] Among them, ST represents the structure tensor, G σ is a Gaussian kernel with standard deviation σ, which enhances the robustness of geometric direction estimation. x and g y They are horizontal gradient and vertical gradient respectively; the structure tensor matrix J has two eigenvalues, denoted as λ1 and λ2, which can be obtained by the following formula:

[0020]

[0021] As a local geometric structure feature descriptor, the eigenvalues ​​of the image structure tensor satisfy the following properties for each pixel:

[0022] If the pixel belongs to a flat area, λ1≈λ2≈0;

[0023] If the pixel belongs to the edge area, λ1>>λ2≈0;

[0024] If the pixel belongs to the corner area, λ1>>λ2>>0;

[0025] The infrared weak target is close to the two-dimensional Gaussian distribution in space, and its local image tensor characteristics are similar to the corner points; therefore, the weighting function W is designed based on the eigenvalue, and the mathematical definition is as follows:

[0026]

[0027] The exponential design of the weighting function based on λ2 and the ratio makes the weight itself pay more attention to the isotropic target, effectively improving the sensitivity to the target;

[0028] C. Fusion of structure tensor information and dilation contrast measurement to obtain the final significant SM:

[0029] SM=DCM*WST. (6)

[0030] D. Finally, the real target is obtained according to the adaptive threshold. The threshold Th is mathematically defined as

[0031] as follows:

[0032] Th=m+k*δ. (7)

[0033] Among them, m represents the mean of the SM saliency map, δ represents the standard deviation of the SM saliency map, and k is a hyperparameter;

[0034] Compared with the prior art, the invention is advantageous in that it has good detection performance for small infrared targets (low signal-to-noise ratio SNR<3) in complex backgrounds.

[0035] The main contributions of the present invention are:

[0036] 1. Use the target prior information to design new structural elements and construct expansion contrast to quickly obtain saliency maps;

[0037] 2. Through the local structure tensor of the image, a weighting function that is sensitive to low signal-to-noise ratio is constructed to effectively reduce the false alarm rate;

[0038] 3. The proposed method has good real-time performance. The present invention has good robustness for targets with low signal-to-noise ratio and effectively improves the detection rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flowchart of the method of the present invention.

[0040] Figure 2 It is a schematic diagram of the structural elements of the dilation operation.

[0041] The black markers are set to 1 for elements related to the dilation operation, and the gray and white markers are set to 0, where the gray marker represents the center point of the structural element.

[0042] Figure 3 The visualization diagram of the whole process of the method of the present invention is shown in FIG. DCM is a dilation operation saliency map, WST is a weighted structure tensor saliency map, and the final saliency map SM is obtained by weighting the two. The white box in the diagram is the real target. DETAILED DESCRIPTION

[0043] The present invention discloses an infrared weak small target detection method based on expansion contrast and structure tensor. Figure 1 As shown: It includes the following steps:

[0044] A. Dilation operation of the original image:

[0045]

[0046] Among them, I raw is the original image, ΔB is the structural element, is the expansion operation; I dilation is the image after the dilation operation, and then the difference joint contrast is constructed with the original image to obtain the significant feature map:

[0047]

[0048] DCM is a dilated contrast saliency map.

[0049] B. Calculate the structure tensor of the original image

[0050]

[0051] Among them, ST represents the structure tensor, G σ is a Gaussian kernel with standard deviation σ, which enhances the robustness of geometric direction estimation. x and g y They are horizontal gradient and vertical gradient respectively; the structure tensor matrix J has two eigenvalues, denoted as λ1 and λ2, which can be obtained by the following formula:

[0052]

[0053] As a local geometric structure feature descriptor, the eigenvalues ​​of the image structure tensor satisfy the following properties for each pixel:

[0054] If the pixel belongs to a flat area, λ1≈λ2≈0;

[0055] If the pixel belongs to the edge area, λ1>>λ2≈0;

[0056] If the pixel belongs to the corner area, λ1>>λ2>>0.

[0057] Infrared weak small targets are close to two-dimensional Gaussian distribution in space, and their local image tensor features are similar to corner points. Therefore, the weighting function W is designed based on the eigenvalue, and the mathematical definition is as follows:

[0058]

[0059] The exponential design of the weighting function based on λ2 and the ratio makes the weight itself pay more attention to isotropic targets, effectively improving the sensitivity to the target.

[0060] C. Fusion of structure tensor information and dilation contrast measurement to obtain the final significant SM:

[0061] SM=DCM*WST. (6)

[0062] D. Finally, the real target is obtained according to the adaptive threshold. The threshold Th is mathematically defined as

[0063] as follows:

[0064] Th=m+k*δ. (7)

[0065] Among them, m represents the mean of the SM saliency map, δ represents the standard deviation of the SM saliency map, and k is a hyperparameter.

[0066] The technical scheme in the embodiment of the present invention will be described in detail below in conjunction with the accompanying drawings in the embodiment of the present invention. Several parameters are involved, and these parameters need to be adjusted according to the specific processing environment to achieve good performance. Based on the embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0067] The test images used in the present invention come from a dataset of small and weak aircraft target detection and tracking in infrared sequence images of the National University of Defense Technology.

[0068] Simulation environment: Matlab2024b;

[0069] Test image: medium-wave infrared image, resolution 256×256, background near the ground;

[0070] Target information: UAV.

[0071] Test steps:

[0072] like Figure 1 As shown, first, the original infrared image is Figure 2 The structural element is expanded, and then the contrast is constructed to obtain the expanded contrast saliency map DCM; then, the structural tensor weight function is constructed to obtain WST, and the two are fused to obtain the final saliency map SM. Finally, the adaptive threshold segmentation is performed to obtain the real target. Its usage effect is as follows Figure 3 shown.

[0073] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for detecting small infrared targets based on expansion contrast and structure tensor, characterized in that: It includes the following steps: First, a dilation template is designed to quickly calculate the dilation contrast; Secondly, the local structure tensor prior of the image is used to design the target representation and construct the weight function to amplify the difference in the local change rate of the target and enhance the target; Finally, the two are weighted to obtain the final saliency map.

2. The infrared small target detection method based on expansion contrast and structure tensor according to claim 1 is characterized in that: The specific steps include: A. Dilation operation of the original image: Among them, I raw is the original image, ΔB is the structural element, is the expansion operation; I dilation is the image after the dilation operation, and then the difference joint contrast is constructed with the original image to obtain the significant feature map: DCM is the dilated contrast saliency map; B. Calculate the local structure tensor of the original image Among them, ST represents the structure tensor, G σ is a Gaussian kernel with standard deviation σ, which enhances the robustness of geometric direction estimation. x and g y They are horizontal gradient and vertical gradient respectively; the structure tensor matrix J has two eigenvalues, denoted as λ1 and λ2, which can be obtained by the following formula: As a local geometric structure feature descriptor, the eigenvalues ​​of the image structure tensor satisfy the following properties for each pixel: If the pixel belongs to a flat area, λ1≈λ2≈0; If the pixel belongs to the edge area, λ1>>λ2≈0; If the pixel belongs to the corner area, λ1>>λ2>>0; The infrared weak target is close to the two-dimensional Gaussian distribution in space, and its local image tensor characteristics are similar to the corner points; therefore, the weighting function W is designed based on the eigenvalue, and the mathematical definition is as follows: The weighting function is designed with the exponential ratio of λ2 to the two eigenvalues, which makes the weight itself pay more attention to the isotropic target and effectively improves the sensitivity to the target; C. Fusion of structure tensor information and dilation contrast measurement to obtain the final significant SM: SM=DCM*WST. (6) D. Finally, the real target is obtained according to the adaptive threshold. The threshold Th is mathematically defined as follows: Th=m+k*δ. (7) Among them, m represents the mean of the SM saliency map, δ represents the standard deviation of the SM saliency map, and k is a hyperparameter.

Images