Infrared small target fast detection method based on gray drop angle contrast algorithm

Abstract

The present application belongs to the field of image processing calculation, and particularly relates to an infrared small target fast detection method based on a gray drop angle contrast algorithm. The method comprises the following steps: S1: acquiring a to-be-detected image, setting a sliding window, and scanning the to-be-detected image by using the sliding window; S2: taking the product of an evaluation result, a difference calculation result and a noise reduction factor as a gray drop angle contrast value; and S3: calculating the gray drop angle contrast value corresponding to the sliding window of each scanning position, taking the region with the gray drop angle contrast value greater than zero as the region where the infrared small target is located, and realizing the fast detection of the infrared small target. The present application provides a method balancing and taking into account excellent detection capability and excellent real-time performance, and can effectively resist noise.

Description

Technical Field

[0001] This invention belongs to the field of image processing and computation, and particularly relates to a fast infrared small target detection method based on grayscale descent angle contrast algorithm. Background Technology

[0002] Infrared small target detection methods have advantages such as good real-time performance, strong detection capability, good portability, ability to detect radar blind spots, strong ability to penetrate smoke and dust, and the ability to work day and night. They also have the characteristics of high detection rate, low false alarm rate, and fast detection speed, making them widely used in civilian and military fields, such as precision guidance, surveillance and remote sensing, and can effectively detect and identify small targets such as drones.

[0003] In recent years, researchers have incorporated the Human Visual System (HVS) into the study of infrared small target detection. The HVS is a hierarchical visual processing system composed of an optical system, a retina, and visual perception pathways. The HVS can quickly locate salient targets in complex external scenes. Although infrared targets are extremely small, they still exhibit certain differences relative to the local background. HVS can utilize these differences to quickly locate these areas. Therefore, the mainstream HVS algorithms are mainly based on local contrast methods: Among them, the Average Absolute Gray Difference (AAGD) algorithm uses a structure of two nested windows to calculate contrast by calculating the difference between the average values ​​of the inner and outer windows, which has low computational complexity and is suitable for practical applications; the Relative Local Contrast (RLCM) method considers the gray difference between the target and the adjacent background, and calculates contrast by sorting the gray values ​​around each pixel to enhance the target or suppress the background; the Three-Layer Template Local Difference Measurement (TTLDM) method is proposed, which combines gray-level difference measurement and variance difference measurement to achieve target enhancement and background suppression for infrared targets; and the Infrared Small Target Detection (ELUM) method based on local component uncertainty measurement and consistency evaluation performs target detection by measuring the uncertainty of local components and signal enhancement based on energy weighting functions. However, these methods still face significant challenges in balancing excellent detection capabilities and real-time performance for infrared small target detection in complex backgrounds. Furthermore, the impact of noise on detection performance is also a problem that cannot be ignored. Summary of the Invention

[0004] In view of this, the present invention aims to provide a fast infrared small target detection method based on grayscale descent angle contrast algorithm, in order to solve the problem that the existing technology cannot balance excellent detection capability and good real-time performance in infrared small target detection under complex backgrounds. The present invention provides a method that balances excellent detection capability and good real-time performance, and can effectively resist noise.

[0005] To achieve the above objectives, the technical solution created by this invention is implemented as follows:

[0006] A fast infrared small target detection method based on grayscale descent angle contrast algorithm, specifically including the following steps:

[0007] S1: Acquire the image to be detected, set a sliding window, and use the sliding window to scan the image to be detected;

[0008] S2: Calculate the sliding window at each scanning position using the energy residual function to pre-evaluate the target area of ​​each sliding window. ;

[0009] The minimum difference between the center pixel and the pixels in the two-layer local neighborhood of the sliding window at each scan position is calculated by minimizing the difference function, thus obtaining the difference calculation results for each sliding window. ;

[0010] Calculate the double gray-scale descent angle corresponding to the sliding window at each scanning position, and calculate the tangent quotient of the double gray-scale descent angle corresponding to the sliding window at each scanning position based on the double gray-scale descent angle. Based on the tangent quotient of the double gray-scale descent angle, obtain the noise reduction factor corresponding to the sliding window at each scanning position.

[0011] Evaluation results , Difference calculation results and noise reduction factor The product of these values ​​is used as the grayscale descent angle contrast value;

[0012] S3: Calculate the gray-scale descent angle contrast value corresponding to the sliding window at each scanning position, and take the area with a gray-scale descent angle contrast value greater than zero as the area where the infrared small target is located, so as to realize the rapid detection of infrared small targets.

[0013] Furthermore, the images to be detected include the GOT-10k dataset.

[0014] Furthermore, in step 2, the energy residual of the current sliding window is calculated using the following formula to pre-evaluate the target region of the current sliding window:

[0015] ;

[0016] ;

[0017] ;

[0018] in, This is the grayscale matrix reconstructed using the average pixel value of the current sliding window. This is a DOG filter, where G is a normally distributed two-dimensional Gaussian function, and x and y are the abscissa and ordinate of the two-dimensional coordinate system in the two-dimensional Gaussian function, respectively. and All values ​​are standard deviations.

[0019] Furthermore, in step S2, when the target size is n×n, the size of the inner local neighborhood pixels is (n+2)×(n+2), and the size of the outer local neighborhood pixels is (n+4)×(n+4).

[0020] Furthermore, in step 2, the minimum difference between the center pixel of the current sliding window and the pixels in the two-layer local neighborhood is calculated using the following formula:

[0021] ;

[0022] in, For the difference calculation results, The grayscale value of the center pixel. The image to be detected. For the inner layer local neighborhood pixels, represents the outer local neighborhood pixels, and ⊕ represents the dilation operation.

[0023] Furthermore, in step S2, the dual gray-scale descent angle of the current sliding window includes the first-order neighborhood gray-scale descent angle and the second-order neighborhood gray-scale descent angle; the set of pixels corresponding to the 8 neighboring pixels adjacent to the center pixel of the current sliding window is defined as the first-order neighborhood, and any local feature in the current sliding window other than the center pixel and the first-order neighborhood is defined as the second-order neighborhood, and the size of the second-order neighborhood is larger than the target size n×n.

[0024] Furthermore, in step S2, the noise reduction factor is calculated. Includes the following steps:

[0025] Calculate the gray-level descent angle of the first-order neighborhood:

[0026] ;

[0027] in, The first-order neighborhood gray-level descent angle, This is the difference between the grayscale value of the center pixel within the current sliding window and the maximum grayscale value of the pixels in its first-order neighborhood. This represents the grayscale value of the center pixel within the current sliding window. The maximum grayscale value of the pixel in the first-order neighborhood;

[0028] Calculate the gray-level descent angle of the second-order neighborhood:

[0029] ;

[0030] in, The gray-level decrease angle of the second-order neighborhood. This is the difference between the grayscale value of the center pixel within the current sliding window and the maximum grayscale value of the pixels in its second-order neighborhood. This represents the grayscale value of the center pixel within the current sliding window. The maximum grayscale value of the pixel in the second-order neighborhood;

[0031] Calculate the tangent quotient of the double gray-level descent angle of the current sliding window based on the first-order and second-order neighborhood gray-level descent angles. :

[0032] , ;

[0033] Based on the dual gray-scale descent angle tangent Calculate the noise reduction factor of the current sliding window:

[0034] .

[0035] Furthermore, the size of the sliding window is larger than the target size.

[0036] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0037] This invention presents a fast infrared small target detection method based on a gray-level descent angle contrast algorithm. It calculates the energy residual of the image using an energy residual function and performs two target localization operations using a minimum difference function, ensuring the reliability of the target information. Through the structural characteristics of the dual gray-level descent angle detection model, it distinguishes the target from salt noise and further enhances the target while suppressing the background, greatly ensuring the accuracy of target detection. Furthermore, it calculates the dual gray-level descent angle tangent quotient in the layer ring regions of the first and second-order neighborhoods, significantly improving computational efficiency, reducing computational complexity, maintaining excellent detection capabilities, and possessing resistance to salt noise interference. Attached Figure Description

[0038] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0039] Figure 1 A flowchart illustrating the rapid infrared small target detection method based on grayscale descent angle contrast algorithm described in the embodiments of the present invention;

[0040] Figure 2A schematic diagram of the grayscale descent angle contrast algorithm described in the embodiment of the present invention;

[0041] Figure 3 A schematic diagram of the grayscale descent angle as described in the embodiments of the present invention;

[0042] Figure 4 A schematic diagram of the dual grayscale descent angle model described in the embodiment of the present invention;

[0043] Figure 5 This is a schematic diagram of the grayscale matrix slicing structure of the target region under the premise that the centroid of the target region is larger than the neighborhood of the target region, as described in the embodiment of the present invention.

[0044] Figure 6 A schematic diagram of the grayscale matrix slice structure of the target region, provided that the centroid of the target region is equal to the neighborhood of the target region, as described in the embodiment of the present invention;

[0045] Figure 7 A schematic diagram of the grayscale matrix slice structure of the noise region as described in the embodiment of the present invention;

[0046] Figure 8 A schematic diagram of the grayscale matrix slice structure of the target outer edge region as described in the embodiment of the present invention;

[0047] Figure 9 This is a schematic diagram of the grayscale matrix slice structure of the continuous strong edge region described in the embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0049] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0050] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0051] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0052] The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0053] like Figures 1-2 As shown, this invention proposes a fast infrared small target detection method based on a gray-scale descent angle contrast algorithm, including: S1: acquiring the image to be detected, setting a sliding window, and scanning the image to be detected using the sliding window; S2: calculating the sliding window at each scanning position using the energy residual function, and pre-evaluating the target area of ​​each sliding window. The difference calculation results for each sliding window are obtained by minimizing the difference function to calculate the minimum difference between the center pixel of the sliding window and the pixels in the two-layer local neighborhood at each scanning position. Calculate the double gray-level descent angle corresponding to the sliding window at each scanning position, and calculate the tangent quotient of the double gray-level descent angle corresponding to the sliding window at each scanning position based on the double gray-level descent angle. Based on the tangent quotient of the double gray-level descent angle, obtain the noise reduction factor corresponding to the sliding window at each scanning position; evaluate the results. , Difference calculation results and noise reduction factor The product of the two is used as the gray-scale descent angle contrast value; S3: Calculate the gray-scale descent angle contrast value corresponding to the sliding window at each scanning position, and take the area with a gray-scale descent angle contrast value greater than zero as the area where the infrared small target is located, so as to realize the rapid detection of infrared small targets.

[0054] It should be noted that this invention proposes a Dual Gray-Level Descending Angle Contrast Algorithm (DGDACM) to simultaneously enhance the target and adaptively suppress complex clutter and salt noise. DGDACM mainly consists of three parts. First, based on the inconsistency between the target and background in the local gray-level neighborhood, this invention calculates the energy residual of the original image using an energy residual function to pre-evaluate potential target regions. Second, based on the gray-level characteristics of the local neighborhood, the minimum difference function is used to calculate the minimum difference between the center pixel and the local neighborhood, and the hierarchical difference between the center pixel and its neighborhood is calculated to enhance and re-localize potential target regions while suppressing background regions. Finally, a noise-resistant detection model based on dual gray-level descending angles further enhances the target and further suppresses smooth background regions, target outer neighborhoods, continuous strong edges, and salt noise regions. Finally, adaptive threshold segmentation is used to obtain accurate target detection results.

[0055] In some instances, the images to be detected typically include the GOT-10k dataset.

[0056] It should be noted that in practical applications, multiple templates of different sizes are defined to calculate the image to be detected, and finally the optimal solution is taken to obtain the infrared weak target.

[0057] In some instances, in step 2, the target region of the current sliding window is pre-evaluated by calculating the energy residual of the current sliding window using the following formula:

[0058] ;

[0059] ;

[0060] ;

[0061] in, This is the grayscale matrix reconstructed using the average pixel value of the current sliding window. This is a DOG filter, where G is a normally distributed two-dimensional Gaussian function, and x and y are the abscissa and ordinate of the two-dimensional coordinate system in the two-dimensional Gaussian function, respectively. and All values ​​are standard deviations.

[0062] It should be noted that, based on the radiation characteristics of small infrared targets, within the local neighborhood of an infrared target, the grayscale value of the central pixel is the largest, and the grayscale change between the central pixel and its central neighborhood is not significant. The size is relatively small, while the inner and outer edges of small infrared targets show drastic changes in grayscale. It allows for a preliminary assessment of potential target areas.

[0063] In some instances, in step S2, when the target size is n×n, the inner local neighborhood pixels ( Figure 1 The size of the inner structure (abbreviated as inner structure) is (n+2)×(n+2), and the size of the outer local neighborhood pixels ( Figure 1 The size of the outer structure (abbreviated as the middle layer) is (n+4)×(n+4).

[0064] It should be noted that, based on the grayscale characteristics of infrared small targets, the target centroid is the maximum grayscale value in its local neighborhood, which must be greater than the grayscale values ​​of all pixels in the local neighborhood. Therefore, this invention can enhance the target and suppress the background by calculating the difference between each pixel and the maximum pixel in different layers.

[0065] In some instances, in step 2, the minimum difference between the center pixel of the current sliding window and the pixels in the two-layer local neighborhood is calculated using the following formula:

[0066] ;

[0067] in, For the difference calculation results, The grayscale value of the center pixel. The image to be detected. For the inner layer local neighborhood pixels, represents the outer local neighborhood pixels, and ⊕ represents the dilation operation.

[0068] It should be noted that, The result is the maximum pixel value of the local neighborhood pixels in the inner layer. The result is the maximum pixel value of the local neighboring pixels in the outer layer.

[0069] Furthermore, H(x) is a step function, defined as follows:

[0070] ;

[0071] Through calculation This enhances the target region because the difference between the pixel center and the maximum pixel value between different layers is positive, while suppressing other regions such as the background because the difference between the pixel center and the maximum pixel value between different layers is negative. It can effectively enhance the target while suppressing the background, enabling secondary localization of the target area and eliminating most background interference.

[0072] In some instances, in step S2, the dual gray-scale descent angle of the current sliding window includes a first-order neighborhood gray-scale descent angle and a second-order neighborhood gray-scale descent angle; the set of pixels corresponding to the 8 neighboring pixels adjacent to the center pixel of the current sliding window is defined as the first-order neighborhood, and any local feature in the current sliding window other than the center pixel and the first-order neighborhood is defined as the second-order neighborhood, and the size of the second-order neighborhood is slightly larger than the target size n×n.

[0073] It should be noted that the set of pixels corresponding to the 8 neighboring pixels adjacent to the center pixel is defined as the first layer of local features, the set of pixels corresponding to the 16 neighboring pixels adjacent to the 8 neighboring pixels is defined as the second layer of local features, and so on. The principle of taking any layer of local features here is the same as above.

[0074] Furthermore, based on the relationship between the center pixel of the current sliding window and its first-order and second-order neighbors, the gray-level descent angles of the first-order and second-order neighbors are calculated. Here, the gray-level descent angle is explained as follows: Figure 3 As shown, in a 3D grayscale image, the grayscale direction of the pixel at the center position is considered as a vertical side (e.g., ...). Figure 3 (As shown by the red dashed line in the middle), the line connecting the central pixel to the pixel corresponding to the maximum grayscale value in its first-order neighborhood is taken as a hypotenuse (e.g., Figure 3 The included angle (not obtuse, as shown by the green dashed line) formed by the two sides (as shown by the green dashed line) Figure 3 As shown by the purple arc in the middle), it is defined as the grayscale decrease angle, and when this angle exceeds R / 2, there will be no decreasing trend. Therefore, in the definition, the present invention sets the grayscale of pixels at the positions of these angles that do not conform to the definition to 0.

[0075] In some instances, the noise reduction factor is calculated. Includes the following steps:

[0076] Calculate the gray-level descent angle of the first-order neighborhood:

[0077] ;

[0078] in, The first-order neighborhood gray-level descent angle, This is the difference between the grayscale value of the center pixel within the current sliding window and the maximum grayscale value of the pixels in its first-order neighborhood. This represents the grayscale value of the center pixel within the current sliding window. The maximum grayscale value of the pixel in the first-order neighborhood;

[0079] Calculate the gray-level descent angle of the second-order neighborhood:

[0080] ;

[0081] in, The gray-level decrease angle of the second-order neighborhood. This is the difference between the grayscale value of the center pixel within the current sliding window and the maximum grayscale value of the pixels in its second-order neighborhood. This represents the grayscale value of the center pixel within the current sliding window. The maximum grayscale value of the pixel in the second-order neighborhood;

[0082] Calculate the tangent quotient of the double gray-level descent angle of the current sliding window based on the first-order and second-order neighborhood gray-level descent angles. :

[0083] , ;

[0084] Based on the dual gray-scale descent angle tangent Calculate the noise reduction factor of the current sliding window:

[0085] .

[0086] It should be noted that the calculation method for the double grayscale descent angle is as follows: Figure 4 As shown, the red area represents the center of the current detection location, the blue area represents the first-order neighborhood, and the yellow area represents the second-order neighborhood. The grayscale decreasing triangle is shown below. Figure 4 As shown, the red lines represent the first-order and second-order vertical edges, the green lines represent the first-order and second-order hypotenuses, the blue lines represent the maximum grayscale value of the first-order neighborhood, the yellow lines represent the maximum grayscale value of the second-order neighborhood, and the purple arcs represent the grayscale descent angles of the first-order and second-order neighborhoods. Furthermore, regarding the selection of template size: the template size for the first-order neighborhood is fixed at 3×3, while the template size for the second-order neighborhood only needs to be equal to or slightly larger than the target size.

[0087] In some embodiments, the size of the sliding window is larger than the target size.

[0088] It should be noted that in the calculation of step S2, the evaluation result is calculated. , Difference calculation results The sliding window size used in this invention is usually the same as that used in the noise reduction factor DGDγ. However, even with different sliding window sizes, the method of this invention can still achieve rapid detection of small infrared targets. For example, calculating the evaluation results... At this time, the size of the sliding window is determined by the center window. The optimal size of the center window must encompass the target signal. The complete minimum window structure has a center window size of 3×3, and the size is adjustable. Adjusting the size of the sliding window can effectively handle targets of different sizes and shapes.

[0089] The saliency of targets is detected based on the infrared radiation characteristics of small infrared targets. Image areas such as sand, stones, soil, or hills are classified as background. The brightness of these backgrounds depends on their ability to radiate infrared waves and is related to their material. Generally, the background is usually a dark area. Some backgrounds with strong infrared radiation appear as brighter areas, but unlike the grayscale features of small targets that are highlighted in a local neighborhood (greater than 1 pixel), the brighter background areas are generally large and continuous. In addition, the grayscale features of salt noise are only highlighted within 1 pixel. Therefore, this invention uses real local grayscale matrix slices as examples to qualitatively analyze the detection of DGDACM algorithm in different regions, discusses the performance of DGDACM algorithm in different detection regions, and demonstrates the effectiveness of DGDACM algorithm.

[0090] Because of I ER Similar to a Gaussian difference filter, it can perform preliminary screening of local region boundaries, identify potential targets within the local area, and then preserve and enhance these boundary regions. Therefore, it enhances target regions, noise regions, the outer edges of target regions, and the outer edges of continuously highlighted regions, while suppressing continuously highlighted inner regions and slowly changing background regions.

[0091] (1) Target region (the target centroid is larger than the centroid neighborhood within the target region)

[0092] When the gray value of the target centroid is greater than some gray values ​​within the target area (i.e., the centroid is the only maximum gray value), such as Figure 5 As shown, since the target area has boundaries, therefore I ER It has been enhanced.

[0093] ;

[0094] In the target area, I MD It has also been enhanced.

[0095] = = ;

[0096] At this point, ∆h1>0, the gray level decreases by an angle in the first-order neighborhood. In the interval (0, π / 4), the tangent value is 1 / 16, which is a constant greater than 0.

[0097] = = ;

[0098] And the second-order descent angle Also in the interval (0, π / 4), the tangent value is 1 / 158, which is a constant greater than 0.

[0099] = ;

[0100] The tangent quotient of the dual gray-scale descent angle is greater than 1. DGDγ is also greater than 1. Therefore, DGDACM effectively enhances the target region.

[0101] (2) Target region (the target centroid is equal to the centroid neighborhood within the target region)

[0102] When the target centroid gray value equals some gray values ​​within the target area (i.e., the centroid is not a unique maximum gray value), such as Figure 6 As shown, the target region has a boundary, thus the IER is enhanced.

[0103] ;

[0104] In the target area, I MD It has also been enhanced;

[0105] = = ;

[0106] At this point, ∆h1=0, and the gray level decreases by an angle in the first-order neighborhood. On the interval (0, π / 2), the tangent value tends towards +∞.

[0107] = = ;

[0108] And the second-order descent angle On the interval (0, π / 4), the tangent value is 1 / (158), which is a constant greater than 0.

[0109] = ;

[0110] The tangent quotient of the double gray-level descent angle is much greater than 1. DGDγ is enhanced. Therefore, the DGDACM algorithm effectively enhances the target region.

[0111] (3) Noise area

[0112] When the testing center happens to be located in the noisy area, such as Figure 7 As shown, the noise region has a boundary, therefore I ER It has been enhanced.

[0113] ;

[0114] In the noisy region, I MD It has also been enhanced.

[0115] = = ;

[0116] At this point, ∆h1>0, the first-order gray-level decrease angle In the interval (0, π / 4), the tangent value is 1 / 138, which is a constant greater than 0.

[0117] = = ;

[0118] And the second-order descent angle It is also in the interval (0, π / 4), with a tangent value of 1 / 128, which is a constant greater than 0.

[0119] = ;

[0120] Due to the grayscale characteristics of noise, the descent angles in the two-order neighborhoods are similar, and the tangent quotient of the double grayscale descent angle is approximately equal to 1. In the noise region, although I ER and I MD While the noise is amplified, DGDγ is approximately zero, and DGDACM is also approximately zero. Therefore, DGDACM effectively suppresses the noise region.

[0121] (4) The outer edge of the target area

[0122] In the local grayscale matrix of the outer edge of the planar target, as shown... Figure 8 As shown, the white dashed line area represents the current pixel center position. The area within the white solid line of the first-level neighborhood is the first-order neighborhood, and the area within the white solid line of the third-level neighborhood is the second-order neighborhood. At this point, ∆h1 < 0, and the gray-level decrease angle of the first-order neighborhood is an obtuse angle. By definition, the tangent quotient (γ) of the double gray-level decrease angle is equal to 0. A boundary exists at the outer edge of the target area; therefore, I... ER It has been enhanced.

[0123] IMD=H[(74-255)]·H((74-151))=0(22);

[0124] At the outer edge of the target region, IMD is suppressed, at which point ∆h1<0, and the gray value descent angle of the first-order neighborhood is obtuse. By definition, the tangent quotient (γ) of the gray value descent angle of the second-order neighborhood is equal to 0, DGDγ is equal to 0, and DGDACM is also equal to 0. Therefore, DGDACM effectively suppresses the outer edge of the target region.

[0125] (5) Continuous strong edge region

[0126] In a local gray-level matrix with continuous strong edges in a plane, when ∆h1 < 0, by definition, the tangent quotient (γ) of the double gray-level descent angle is equal to 0. Therefore, we only need to discuss the case where ∆h1 ≥ 0. Based on the gray-level characteristics of continuous strong edges, the gray-level change is not significant in the local region, and the gray-level difference is approximately 0; therefore, we can consider ∆h1 ≈ 0. Thus, we will consider the case where ∆h1 = 0 for the gray-level matrix (…). Figure 9 We will discuss matrices containing grayscale values, such as... Figure 9 As shown.

[0127] There is a boundary within a continuous strong edge region, but the detection center is within a continuous highlight edge region; therefore, I ER It is suppressed.

[0128] ;

[0129] In regions with continuous strong edges, IMD is suppressed.

[0130] = = ;

[0131] In the detection region of continuous strong edges, since ∆h1=0, the gray level decrease angle of the first-order neighborhood tends to π / 2+, and the tangent value tends to positive infinity.

[0132] = = ;

[0133] At the same time, ∆h2=0, the second-order gray level decrease angle also tends to π / 2+, and the tangent value tends to positive infinity.

[0134] = ;

[0135] At this time, the tangent quotient When x approaches π / 2+, it is equivalent to infinity, therefore γ=1, DGDγ equals 0, and DGDACM also equals 0. Therefore, DGDACM effectively suppresses continuous strong edge regions.

[0136] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0137] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A fast infrared small target detection method based on grayscale descent angle contrast algorithm, characterized in that: Specifically, the steps include the following: S1: Acquire the image to be detected, set a sliding window, and use the sliding window to scan the image to be detected; S2: Calculate the sliding window at each scanning position using the energy residual function to pre-evaluate the target area of ​​each sliding window. ; In step 2, the energy residual of the current sliding window is calculated using the following formula to pre-evaluate the target region of the current sliding window: ； ； ； in, This is the grayscale matrix reconstructed using the average pixel value of the current sliding window. This is a DOG filter, where G is a normally distributed two-dimensional Gaussian function, and x and y are the abscissa and ordinate of the two-dimensional coordinate system in the two-dimensional Gaussian function, respectively. and All are standard deviations; The minimum difference between the center pixel and the pixels in the two-layer local neighborhood of the sliding window at each scan position is calculated by minimizing the difference function, thus obtaining the difference calculation results for each sliding window. ; Calculate the double gray-scale descent angle corresponding to the sliding window at each scanning position, and calculate the tangent quotient of the double gray-scale descent angle corresponding to the sliding window at each scanning position based on the double gray-scale descent angle. Based on the tangent quotient of the double gray-scale descent angle, obtain the noise reduction factor corresponding to the sliding window at each scanning position. Evaluation results , difference calculation results and noise reduction factor The product of these values ​​is used as the grayscale descent angle contrast value; S3: Calculate the gray-scale descent angle contrast value corresponding to the sliding window at each scanning position, and take the area with a gray-scale descent angle contrast value greater than zero as the area where the infrared small target is located, so as to realize the rapid detection of infrared small targets.

2. The method for rapid detection of small infrared targets based on grayscale descent angle contrast algorithm according to claim 1, characterized in that: The images to be detected include the GOT-10k dataset.

3. The method for rapid detection of small infrared targets based on grayscale descent angle contrast algorithm according to claim 1, characterized in that: In step S2, when the target size is n×n, the size of the inner local neighborhood pixels is (n+2)×(n+2), and the size of the outer local neighborhood pixels is (n+4)×(n+4).

4. The method for rapid detection of small infrared targets based on grayscale descent angle contrast algorithm according to claim 3, characterized in that: In step 2, the minimum difference between the center pixel of the current sliding window and the pixels in the two-layer local neighborhood is calculated using the following formula: ； in, For the difference calculation results, The grayscale value of the center pixel. The image to be detected. For the inner layer local neighborhood pixels, represents the outer local neighborhood pixels, and ⊕ represents the dilation operation.

5. The method for rapid detection of small infrared targets based on grayscale descent angle contrast algorithm according to claim 1, characterized in that: In step S2, the dual gray-scale descent angle of the current sliding window includes the first-order neighborhood gray-scale descent angle and the second-order neighborhood gray-scale descent angle; the set of pixels corresponding to the 8 neighboring pixels adjacent to the center pixel of the current sliding window is defined as the first-order neighborhood, and any local feature in the current sliding window other than the center pixel and the first-order neighborhood is defined as the second-order neighborhood, and the size of the second-order neighborhood is larger than the target size n×n.

6. The method for rapid detection of small infrared targets based on grayscale descent angle contrast algorithm according to claim 5, characterized in that: In step S2, the noise reduction factor is calculated. Includes the following steps: Calculate the gray-level descent angle of the first-order neighborhood: ； in, The first-order neighborhood gray-level descent angle, This is the difference between the grayscale value of the center pixel within the current sliding window and the maximum grayscale value of the pixels in its first-order neighborhood. This represents the grayscale value of the center pixel within the current sliding window. The maximum grayscale value of the pixel in the first-order neighborhood; Calculate the gray-level descent angle of the second-order neighborhood: ； in, The gray-level decrease angle of the second-order neighborhood. This is the difference between the grayscale value of the center pixel within the current sliding window and the maximum grayscale value of the pixels in its second-order neighborhood. This represents the grayscale value of the center pixel within the current sliding window. The maximum pixel grayscale value in the second-order neighborhood; Calculate the tangent quotient of the double gray-level descent angle of the current sliding window based on the first-order and second-order neighborhood gray-level descent angles. : , ； Based on the dual gray-scale descent angle tangent Calculate the noise reduction factor of the current sliding window: 。 7. The method for rapid detection of small infrared targets based on grayscale descent angle contrast algorithm according to claim 1, characterized in that: The size of the sliding window is larger than the target size.

Images