A method for infrared small target detection based on image curvature and gradient
Through the detection method based on image curvature and gradient, combined with curvature filtering and gradient filtering, centroid point screening is performed after fusion, the problem of high false alarm rate in infrared small object detection technology is solved, the detection accuracy and performance are improved, and the complex environment is adapted to The detection requirements for infrared small targets.
Patent Information
- Application Number
- CN202111116243.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-23
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2041-09-23
AI Technical Summary
The existing infrared small-objective detection technology has shortcomings in the distinction between small-objective feature extraction and background clutter, resulting in high false alarm rates and leaked alarm rates, making it difficult to improve the working distance and practical application capabilities of the infrared detection system.
The detection method based on image curvature and gradient is adopted, and the minimum curvature and minimum gradient feature maps are obtained through curvature filtering and gradient filtering. After fusing, the infrared image is subject to centroid point filtering to reduce false alarm rate and suppress background clutter.
It improves the accuracy and performance of infrared small target detection, reduces false alarm rate, adapts to the demand for infrared small target detection in complex environments, and has high engineering application value.
Smart Images

Figure CN113776676B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of infrared detection, and in particular relates to an infrared small target detection method based on image curvature and gradient. Background Art
[0002] Infrared detection technology relies on the infrared radiation characteristics of the receiving target to perform passive detection. It combines photoelectric sensing technology, image processing technology, and control technology. Even in harsh electromagnetic environments, it can implement day and night detection of ground targets, air targets, and even space targets. The infrared detection system is small in size and light in weight, and can be flexibly deployed in various terrains. In addition, it has the advantages of strong maneuverability, good anti-interference performance, and the ability to work day and night. The role of infrared target detection technology in military or civilian fields is crucial.
[0003] Infrared target detection, especially infrared small target detection technology, as an indispensable key technology in the fields of infrared imaging guidance, infrared warning, infrared search and tracking, has become a hot research topic in the field of infrared image processing. Since the improvement of the effective range, an important indicator in the infrared detection system, is an urgent problem to be solved, domestic and foreign scholars have proposed the concept of small targets for infrared surveillance systems since the late 1970s: small targets refer to targets that are far away from the infrared imaging system and only occupy a few pixels or even one pixel in the image. Infrared small target detection technology is a key technology to improve the effective range of the infrared detection system, and the analysis of the characteristics of small targets themselves is a crucial research to solve this problem.
[0004] Existing infrared small target detection technology often filters the data obtained by the detector to suppress the clutter and noise in the background as much as possible, and extracts the pixel points exceeding the threshold as the target observation value for each frame of the image with a fixed or adaptive threshold. However, due to the small number of pixels occupied by small targets on the detector plane and the unclear texture features, coupled with the existing domestic infrared small target detection technology still has great deficiencies in the extraction of small target features and the distinction from background clutter, the infrared small target detection technology has a high false alarm rate and missed alarm rate in practical applications, that is, the target detection error or failure to detect the target, which greatly reduces the ability of infrared warning or infrared imaging guidance to detect the target, resulting in an embarrassing situation where the combat capability does not meet the standard. Therefore, it is of great significance to analyze the characteristics of infrared small targets by using the difference between the target and the background, and then improve or solve the problem of high false alarm rate in the current small target detection according to the unique characteristics of small targets. Summary of the invention
[0005] The purpose of the present invention is to propose an infrared small target detection method based on image curvature and gradient. The existing infrared small target detection technology still has great deficiencies in the extraction of small target features and the distinction from background clutter, resulting in high false alarm rate and missed alarm rate in practical applications of the infrared small target detection technology, and it is difficult to adapt to the needs of infrared small target detection in complex environments. The detection method has low computational complexity and is simple to implement, and has high engineering application value.
[0006] The specific technical solution of the present invention is: a method for detecting small infrared targets based on image curvature and gradient, comprising the following steps:
[0007] Step 1: Input infrared image f I Perform curvature filtering to obtain the minimum curvature feature map f curv_min And the mean curvature feature map f curv_ave .
[0008] Step 2: According to the minimum curvature feature map f curv_min Set the threshold and perform global threshold segmentation on the minimum curvature feature map to obtain the binary map f th .
[0009] Step 3: Combine the binary image f th Perform gradient filtering on the input infrared image to obtain the minimum gradient feature map f grad_min And the average gradient feature map f grad_Ave .
[0010] Step 4: Fuse the minimum curvature feature map and the minimum gradient feature map to obtain the small target feature map F mix .
[0011] Step 5: Combine small target feature map F mix and infrared image f I , for the binary image f th Perform centroid screening and output all valid centroids, which are the suspected target centroids.
[0012] The effects of the present invention are as follows: 1) compared with the traditional infrared small target detection method, the small target characteristic value is obtained according to the fusion of curvature and gradient, and the small target characteristic value is used to screen the effective centroid point, which greatly reduces the false alarm while ensuring the detection rate; 2) compared with the traditional infrared small target detection method, the small target characteristic value is used as a means of screening the effective target, which well suppresses the background clutter, and has better performance advantages in scenes with complex backgrounds such as the ground; 3) the method is a supplement to the existing infrared small target detection method, is easy to implement, and has a more practical engineering application value.
[0013] The present invention is further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 A flow chart of an infrared small target detection method based on image curvature and gradient described in the present invention.
[0015] Figure 2 The present invention discloses a flow chart for screening effective targets based on small target feature maps and grayscale maps.
[0016] Figure 3 Actual scene diagram of UAV infrared target.
[0017] Figure 4 Binary processing result of actual scene of UAV infrared target.
[0018] Figure 5 A diagram of the infrared target detection results of a UAV using the detection method described in the present invention. DETAILED DESCRIPTION
[0019] Combination Figure 1-2 The infrared small target detection method based on image curvature and gradient of the present invention comprises the following steps:
[0020] Step 1: Perform curvature filtering on the input infrared image to obtain the minimum curvature feature map f curv_min And the mean curvature feature map f curv_ave , as follows:
[0021] Step 11: The input infrared image sets three windows as n×n basic filter templates W1, W2, and W3, considering that the size of small targets is less than 7×7, n∈{5,7}.
[0022] Step 12: According to the basic filter templates W1, W2, and W3, the four direction filter templates W0, W1, W2, and W3 are calculated respectively. 45 , W 90 , W 135 .
[0023] Step 13: Calculate the second-order derivative f″ of each pixel in the input infrared image in the four directions of 0°, 45°, 90°, and 135° ij0 , f″ ij45 , f″ ij90 , f″ ij135 , the infrared image resolution is i×j, i is the number of image rows, and j is the number of image columns.
[0024] Step 14: All the above second-order derivatives f″ ij0 , f″ ij45 , f″ ij90 , f″ ij135The positive part of is set to 0, and then its absolute value is taken respectively, so that the curvature characteristic value C of each pixel in four directions is obtained. ij0 , C ij45 , C ij90 , C ij135 .
[0025] Step 15: For each pixel of the infrared image, calculate C ij0 , C ij45 , C ij90 , C ij135 The minimum curvature eigenvalue f ijcurv_min and the mean curvature eigenvalue f ijcurv_ave After all pixel points are calculated, the minimum curvature feature map f of the infrared image is obtained. curv_min And the mean curvature feature map f curv_ave .
[0026] Step 2: According to the minimum curvature feature map f curv_min Set the threshold and perform global threshold segmentation on the minimum curvature feature map to obtain the binary map f th , as follows:
[0027] Step 21: Calculate the minimum curvature feature map f curv_min The global mean μ and variance σ of .
[0028] Step 22: Set the threshold T for threshold segmentation th =μ+k·σ, where the adjustment coefficient k takes an empirical value of 0.1 and can be adjusted based on the binary result.
[0029] Step 23: For the infrared image f I For each pixel in the ijcurv_min With threshold T th If the size is greater than T th If the pixel is not deleted, the binarization result is set to 0.
[0030] Step 24: After all pixel points are compared, all points whose binarization results are 1 are output, and the binarized image f is obtained. th .
[0031] Step 3: Combine the binary image f th Perform gradient filtering on the input infrared image to obtain the minimum gradient feature map f grad_min And the average gradient feature map f grad_Ave , as follows:
[0032] Step 31: Take the binary image f thA pixel with a median value of 1 is taken as the center, and an infrared sub-image area with a window of m×m pixels is obtained. Considering that the size of small targets is less than 7×7, m∈{5,7,9}.
[0033] Step 32: Calculate the grayscale difference between the current pixel and the pixel points in the eight directions of 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315° in the infrared sub-image area until the binary image f is traversed. th All the pixels with a value of 1 in the binarization result are obtained, and the gradient eigenvalues G in eight directions of each pixel with a binarization result of 1 are obtained. ij0 , G ij45 , G ij90 , G ij135 , G ij180 , G ij225 , G ij270 , G ij315 .
[0034] Step 33: Set the gradient feature values in eight directions of the pixel points whose binarization result is 0 to 0.
[0035] Step 34: Calculate G for each pixel in the infrared image ij0 , G ij45 , G ij90 , G ij135 , G ij180 , G ij225 , G ij270 , G ij315 The minimum gradient eigenvalue f ijgrad_min and the average gradient eigenvalue f ijgrad_Ave After all pixel points are calculated, the minimum gradient eigenvalue f of the infrared image is obtained. grad_min and the average gradient eigenvalue F grad_Ave .
[0036] Step 4: Fuse the minimum curvature feature map and the minimum gradient feature map to obtain the small target feature map F mix ,Since the grayscale characteristics of bright small targets are generally bright in the middle and dark around, and the grayscale of the background generally changes slowly and is relatively flat, the curvature and gradient of the small target and the flat background are quite different. The small target feature value obtained by fusing these two values can be better distinguished from the background clutter, as follows:
[0037] Step 41: for each pixel point of the infrared image, determine the curvature characteristic value C in four directions ij0 , C ij45 , C ij90 , C ij135 and the gradient eigenvalues G in eight directions ij0 , Gij45 , G ij90 , G ij135 , G ij180 , G ij225 , G ij270 , G ij315 Are they of the same sign? If they are of the same sign, the curvature features are marked with the same sign h ijc The same sign as the gradient feature h ijg are all set to 1, otherwise h ijc and h ijg All are set to 0, because the curvature and gradient values of small targets in multiple directions are often positive or negative. When it is a bright small target, they are all negative, and when it is a dark small target, they are all positive.
[0038] Step 42: According to the minimum gradient feature map f of the infrared image grad_min and the minimum curvature feature map f curv_min Calculate the global maximum value and obtain the global maximum value f of the minimum curvature feature map of the infrared image curv_min_max and the global maximum value f of the minimum gradient feature map grad_min_max .
[0039] Step 43: for each pixel of the infrared image, according to its minimum curvature eigenvalue f ijcurv_min , mean curvature eigenvalue f ijcurv_ave , minimum gradient eigenvalue f ijgrad_min , average gradient eigenvalue f ijgrad_Ave , the global maximum value f of the minimum curvature feature map curv_min_max , the global maximum value f of the minimum gradient feature map grad_min_max The curvature feature is marked with the same sign as h ijc The same sign as the gradient feature h ijg , calculate the small target eigenvalue F ijmix :
[0040] F ijmix =α×(f ijcurv_min / f curv_min_max )×h ijc +β×(f ijgrad_min / f grad_min_max )×h ijg
[0041] Among them, α and β are adjustment coefficients, α, β∈[0,1], and the default value is 0.5.
[0042] Step 44: the small target feature value F of each pixel of the infrared image ijmix After the calculation is completed, the small target feature map F of the infrared image is obtained. mix .
[0043] Step 5: Combine the small target feature map F of the infrared image mixand infrared image f I For the binary image f th Perform centroid screening and output all valid centroids as test results, as follows:
[0044] Step 51, take the binary image f th A pixel with a median value of 1 is taken as the center, and a binary sub-image area with a window of p×p pixels is obtained. Considering that the size of small targets is less than 7×7, p∈{5,7,9}.
[0045] Step 52: determine whether there are other pixels in the binary sub-image region of p×p pixels. If so, calculate the mean value F of the small target feature values of all pixels in the binary sub-image region. mix_ave , calculate the grayscale mean f of all pixels in the binary sub-image area I_ave , compare the F of the current center pixel ijmix Is the value greater than F? mix_ave If it is greater than, keep it, otherwise delete the pixel, and then determine the gray value f of the current center pixel. ijI Is it greater than f I_ave , if it is greater than, keep it, otherwise delete the pixel.
[0046] Step 53: traverse the binary graph f th Finally, all the filtered valid centroid points are output, which are the suspected target centroid points.
[0047] Example 1
[0048] Combination Figure 3 to Figure 5 , the following describes the embodiments of the present invention through specific examples, such as Figure 3 As shown in the figure, the input resolution is 640×512 for the infrared image of the drone to be detected. There are two drone targets flying in the air, and the ground background is relatively complex, with buildings, roads, trees, etc. The following uses the "infrared small target detection method based on image curvature and gradient" proposed in this patent to detect the small targets in this scene. The implementation steps are as follows:
[0049] Step 1: Perform curvature filtering on the input drone infrared image to obtain the minimum curvature feature map f curv_min And the mean curvature feature map f curv_ave , the basic filter template window size is 5.
[0050] Step 2: According to the minimum curvature feature map f curv_min Set the threshold for threshold segmentation, T th =μ+k·σ, where the adjustment coefficient k takes an empirical value of 0.1, which can be adjusted according to the binary result. Then, the minimum curvature feature map is segmented by global threshold to obtain the binary map f th ,like Figure 4 As shown, it can be seen that the number of points to be detected is greatly reduced without losing the target, and the background clutter is suppressed.
[0051] Step 3: Combine the binary image f th Perform gradient filtering on the input infrared image. Considering that the size of small targets is less than 7×7, m is set to 7, and the minimum gradient feature map f is calculated. grad_min And the average gradient feature map f grad_Ave .
[0052] Step 4: Fuse the minimum curvature feature map and the minimum gradient feature map to obtain the small target feature map F mix .
[0053] Step 5: Combine small target feature map F mix and infrared image f I , for the binary image f th The centroid points are screened. The size of the binarized sub-image area is 7×7, and all valid centroid points are output, such as Figure 5 As shown in the figure, the bright spot is the centroid of the suspected target. It can be seen that compared with the original infrared image, the detection result greatly suppresses the background clutter, reducing the false alarm rate while ensuring the detection rate.
[0054] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.
Claims
1. A method for detecting small infrared targets based on image curvature and gradient, characterized in that: The following steps are involved: Step 1: Input infrared image f I Perform curvature filtering to obtain the minimum curvature feature map f curv_min And the mean curvature feature map f curv_ave , including the following steps: Step 11: Input infrared image f I Set the three windows as n×n basic filter templates W1, W2, W3, n∈{5,7}; Step 12: According to the basic filter templates W1, W2, and W3, the four direction filter templates W0, W1, W2, and W3 are calculated respectively. 45 , W 90 , W 135 ; Step 13: For the infrared image f I For each pixel in the image, calculate the second-order derivative f″ in the four directions of 0°, 45°, 90°, and 135° ij0 , f″ ij45 , f″ ij90 , f″ ij135 , i is the number of image rows, j is the number of image columns; Step 14: All the above second-order derivatives f″ ij0 , f″ ij45 , f″ ij90 , f″ ij135 The positive part of is set to 0, and then its absolute value is taken respectively, so that the curvature characteristic value C of each pixel in four directions is obtained. ij0 , C ij45 , C ij90 , C ij135 ; Step 15: For each pixel of the infrared image, calculate C ij0 , C ij45 , C ij90 , C ij135 The minimum curvature eigenvalue f ijcurv_min and the mean curvature eigenvalue f ijcurv_ave After all pixel points are calculated, the minimum curvature feature map f of the infrared image is obtained. curv_min And the mean curvature feature map f curv_ave ; Step 2: According to the minimum curvature feature map f curv_min Set the threshold and perform global threshold segmentation on the minimum curvature feature map to obtain the binary map f th , including the following steps: Step 21: Calculate the minimum curvature feature map f curv_min The global mean μ and variance σ of ; Step 22: Set the threshold T for threshold segmentation th =μ+k·σ, where the adjustment coefficient k takes the empirical value of 0.1; Step 23: For the infrared image f I For each pixel in the ijcurv_min With threshold T th If the size is greater than T th If the pixel is retained, the binarization result is set to 1; otherwise, the pixel is deleted, and the binarization result is set to 0. Step 24: After all pixel points are compared, all points whose binarization results are 1 are output, and the binarized image f is obtained. th ; Step 3: Combine the binary image f th For infrared image f I Perform gradient filtering to obtain the minimum gradient feature map f grad_min And the average gradient feature map f grad_Ave , including the following steps: Step 31: Take the binary image f th A pixel with a median value of 1, with the current pixel as the center, obtains an infrared sub-image area with a window of m×m pixels, m∈{5,7}; Step 32: Calculate the grayscale difference between the current pixel and the pixel points in the eight directions of 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315° in the infrared sub-image area until the binary image f is traversed. th All the pixels with a value of 1 in the binarization result are obtained, and the gradient eigenvalues G in eight directions of each pixel with a binarization result of 1 are obtained. ij0 , G ij45 , G ij90 , G ij135 , G ij180 , G ij225 , G ij270 , G ij315 ; Step 33, setting the gradient feature values in eight directions of the pixel points whose binarization result is 0 to 0; Step 34: Calculate G for each pixel in the infrared image ij0 , G ij45 , G ij90 , G ij135 , G ij180 , G ij225 , G ij270 , G ij315 The minimum gradient eigenvalue f ijgrad_min and the average gradient eigenvalue f ijgrad_Ave After all pixel points are calculated, the minimum gradient feature map f of the infrared image is obtained. grad_min And the average gradient feature map f grad_Ave; Step 4: According to the minimum curvature feature map f curv_min And the minimum gradient feature map f grad_min Fusion is performed to obtain the small target feature map F mix , including the following steps: Step 41: for each pixel point of the infrared image, determine the curvature characteristic value C in four directions ij0 , C ij45 , G ij90 , G ij135 and the gradient eigenvalues G in eight directions ij0 , G ij45 , G ij90 , G ij135 , G ij180 , G ij225 , G ij270 , G ij315 Are they of the same sign? If they are of the same sign, the curvature features are marked with the same sign h ijc The same sign as the gradient feature h ijg are all set to 1, otherwise h ijc and h ijg All are set to 0; Step 42: According to the minimum gradient feature map f of the infrared image grad_min and the minimum curvature feature map f curv_min Calculate the global maximum value and obtain the global maximum value f of the minimum curvature feature map of the infrared image curv_min_max and the global maximum value f of the minimum gradient feature map grad_min_max ; Step 43: for each pixel of the infrared image, according to the minimum curvature eigenvalue f ijcurv_min , mean curvature eigenvalue f ijcurv_ave , minimum gradient eigenvalue f ijgrad_min , average gradient eigenvalue f ijgrad_Ave , the global maximum value f of the minimum curvature feature map curv_min_max , the global maximum value f of the minimum gradient feature map grad_min_max , the curvature feature is marked with the same sign h ijc The same sign as the gradient feature h ijg , calculate the small target eigenvalue F ijmix ; F ijmix =α×(f ijcurv_min / f curv_min_max )×h ijc +β×(f ijgrad_min / f grad_min_max )×h ijg Among them, α and β are adjustment coefficients, α, β∈[0,1]; Step 44: the small target feature value F of each pixel of the infrared image ijmix After the calculation is completed, the small target feature map F of the infrared image is obtained. mix ; Step 5: Combine small target feature map F mix and infrared image f I , for the binary image f th Perform centroid screening and output all valid centroids, which are the suspected target centroids.
2. The infrared small target detection method based on image curvature and gradient according to claim 1 is characterized in that: In step 5, the small target feature map F of the infrared image is combined mix and infrared image f I For the binary image f th Performing centroid screening and outputting all valid centroids as detection results includes the following steps: Step 51, take the binary image f th A pixel with a median value of 1, with the current pixel as the center, obtains a binary sub-image area with a window of p×p pixels, p∈{5,7}; Step 52: determine whether there are other pixels in the binary sub-image region of p×p pixels. If so, calculate the mean value F of the small target feature values of all pixels in the binary sub-image region. mix_ave , calculate the grayscale mean f of all pixels in the binary sub-image area I_ave , compare the small target feature value F of the current center pixel ijmix Is the value greater than F? mix_ave If it is greater than, keep it, otherwise delete the pixel, and then determine the gray value f of the current center pixel. ijI Is it greater than f I_ave , if it is greater than , keep it, otherwise delete the pixel; Step 53: traverse the binary graph f th Finally, all the filtered valid centroid points are output, which are the suspected target centroid points.
