An infrared dim small target detection method based on adaptive contrast enhancement

Abstract

The application discloses an infrared dim small target detection method based on adaptive contrast enhancement and belongs to the technical field of computer image infrared dim small target detection. The application starts from the background around the target, first fits the gradient field of a local region, and uses the gradient field to construct a target local region divergence feature map to filter out the flat clutter around the target and the pixel size noise under high brightness; then uses a gradient direction template to weaken the background clutter around the target and the target-like background clutter on this basis; through the fusion of the two methods, the noise and clutter around the target can be adaptively filtered out, so that when subsequent local contrast enhancement is carried out, the potential target region can obtain a higher weight to obtain a final result saliency map, the accuracy of target detection is improved, and the anti-aliasing capability is excellent.

Description

Technical Field

[0001] This invention relates to the field of infrared weak target detection technology in computer image processing, and specifically to an infrared weak target detection method based on adaptive contrast enhancement. Background Technology

[0002] Small target detection has applications in various fields, such as active defense, early warning, and maritime rescue. Due to the long optical imaging distance, projected targets in infrared images are typically very small (i.e., less than 9×9 pixels), providing few clues in terms of color, shape, and texture. Furthermore, the infrared radiation of a target decreases significantly after long-range propagation, resulting in a low signal-to-noise ratio (SNR) for small infrared targets. Therefore, in the presence of strong noise and complex background clutter, weak targets with low SNR are easily obscured by the background, leading to missed detections or false alarms. Thus, robust detection of small targets against complex infrared backgrounds remains a challenging task.

[0003] Filter-based methods (i.e., Top-hat filters and MaxMedian / MaxMean filters) detect small targets in the foreground image obtained by subtracting the estimated background image from the input image. While these methods are relatively easy to implement, they are generally only effective on smooth backgrounds. Inspired by the Human Visual System (HVS), local contrast-based methods assume that the target is not the brightest region in the image, but rather the region with the highest contrast. These methods have great potential in terms of detection accuracy and computational speed. Most existing local contrast-based methods directly calculate the contrast between the center and its neighborhood. Even though these methods can achieve good performance on smooth and flat backgrounds, they struggle to detect small targets trapped in complex noise and clutter. To address this, an infrared weak target detection method based on adaptive contrast enhancement is proposed. Summary of the Invention

[0004] The technical problem to be solved by this invention is: how to effectively improve the accuracy of weak targets in complex backgrounds. This invention provides an infrared weak target detection method based on adaptive contrast enhancement. Based on the existing local contrast detection method, this invention further improves the detection efficiency of the algorithm and provides new ideas for subsequent local contrast algorithm research and engineering applications.

[0005] The present invention solves the above-mentioned technical problems through the following technical solution, and the present invention includes the following steps:

[0006] Step S1: Read the input infrared image sequence;

[0007] Step S2: Construct a 5×5 local image region with each pixel as the center point, and use manual fitting to fit the local image region into a continuous bivariate cubic function.

[0008] Step S3: Based on the continuous bivariate cubic function obtained in step S2, calculate the gradient magnitude and gradient direction information at the center point to obtain the gradient vector diagram;

[0009] Step S4: Based on the gradient vector map obtained in step S3, calculate the divergence feature corresponding to each pixel as the center point to obtain the divergence feature map.

[0010] Step S5: Based on the gradient direction information obtained in step S2, construct a gradient direction template to calculate the histogram weights at different positions, and obtain the histogram weight map.

[0011] Step S6: After fusing the divergence feature map and histogram weight map obtained in steps S4 and S5 and processing them, a weighted divergence feature map is obtained.

[0012] Step S7: Calculate the local contrast measurement coefficients at each location of the weighted divergence feature map, apply the coefficients at the corresponding locations to the image to obtain the target saliency map, and finally obtain the target location through adaptive threshold segmentation.

[0013] Furthermore, in step S1, the input infrared image I is read, with an image size of H×W, where W is the width of the image and H is the height of the image.

[0014] Furthermore, in step S2, the fitted continuous bivariate cubic function is as follows:

[0015]

[0016] Among them, the function g i (r,c) is a set of symmetric second-order Chebyshev polynomials obtained by neglecting terms of fourth order and above from the first-order Chebyshev polynomials R and C. To replace, coefficient k i as follows:

[0017]

[0018] Where r and c take values ​​in the range of [-2, 2], corresponding to each point from left to right in the 5×5 local image region.

[0019] Furthermore, in step S3, the specific process of obtaining the gradient vector diagram is as follows:

[0020] Step S31: After obtaining the function f(r,c) for the local image region, calculate the directional derivatives at 0° and 90° corresponding to the center point, denoted as f0.′ and f9 ′ 0:

[0021]

[0022] Where α represents the direction;

[0023] Step S32: Then calculate the gradient magnitude and direction information corresponding to the center point, represented as follows: and

[0024]

[0025] The range of arctan is:

[0026] Step S33: Obtain the gradient vector map G by calculating the gradient magnitude and direction information at each location in the image.

[0027] Furthermore, in step S4, a 5×5 local image region is constructed in the gradient field with each pixel as the center, and the divergence of the center point is Δf(0,0):

[0028]

[0029] The divergence feature map, denoted as D(x,y), can be obtained by calculating the divergence value of each point according to the above formula.

[0030] Furthermore, in step S5, the specific process of obtaining the histogram weight map is as follows:

[0031] Step S51: Since the gradient direction at the target center point is time-varying, the gradient direction value at the center point is deleted, and the remaining matrix is ​​M:

[0032] M(r,c)=del(I c )

[0033] Among them, I c This indicates the center position of a 5×5 local image region, and del represents the function to delete a specific element;

[0034] Step S52: Divide the remaining elements into class b, and calculate the number of elements belonging to the corresponding class by comparing them with the gradient direction template, denoted as N(k):

[0035] N(k) = hist(M(k), b)

[0036] Where M(k) represents the matrix region of the k-th class after deleting the gradient direction value at the center point, b represents the number of calculated classes, hist represents the histogram calculation function, and the gradient direction template represents the direction matrix of all positions in the local image region pointing to the center point.

[0037] Step S53: If the number calculated for the k-th category is greater than... Will be restricted to Conversely, if it is less than The weight of each category region is determined by the actual number of calculations, and is denoted as W(k):

[0038]

[0039] Step S54: Calculate the weight of the center point, denoted as W(x,y):

[0040]

[0041] Where x and y represent the width and height of a pixel in the original infrared image;

[0042] Step S55: Calculate the center point weight W(x,y) with each point in the image as the center point to obtain the histogram weight matrix, which is the histogram weight map.

[0043] Furthermore, in step S6, the specific processing procedure is as follows:

[0044] Step S61: Merge the divergence feature map and histogram weight map obtained in steps S4 and S5 to obtain the fused divergence feature map, denoted as S(x,y):

[0045] S(x,y)=W(x,y)×D(x,y);

[0046] Step S62: Since the divergence of the target center region is less than 0, ignore the positive value part of the fused divergence feature map, and then take the absolute value of all negative values ​​to obtain the absolute value map of the weighted divergence, represented as A(x,y):

[0047] A(x,y) = |min(0,S(x,y))|;

[0048] Step S63: Squaring the absolute value map of the weighted divergence yields the squared weighted absolute value map of divergence, i.e., the weighted divergence feature map, denoted as WD(x,y):

[0049] WD(x,y)=A(x,y)×A(x,y).

[0050] Furthermore, in step S7, the specific processing procedure is as follows:

[0051] Step S71: Construct a 5×5 central region, denoted as region 0, with each pixel in the squared weighted absolute divergence map as the center position. Construct regions of the same size around region 0, denoted as regions 1-8. Calculate the average value of each region:

[0052]

[0053] Where (x,y) represents the element position of the square-weighted absolute divergence plot, i represents the i-th region, i = 1, 2, ... 8, and j represents the element position in the i-th region;

[0054] Step S72: Calculate the average value of each area surrounding the central area, denoted as m. b :

[0055]

[0056] Based on the average value m b The local contrast measurement coefficients are constructed and applied to the squared-weighted absolute divergence map to obtain the target saliency map, denoted as Out(x,y):

[0057]

[0058] Where L represents the maximum value in region 0;

[0059] Step S73: Obtain the target center position through adaptive threshold T segmentation:

[0060] T=μ+k×σ

[0061] Where μ represents the mean of the saliency plot, σ represents the standard deviation of the saliency plot, and k represents the coefficient.

[0062] Compared with existing technologies, this invention has the following advantages: This infrared weak target detection method based on adaptive contrast enhancement starts from the background around the target, first fitting the gradient field of the local region, and using the gradient field to construct a divergence feature map of the target's local region to filter out flat clutter and pixel-sized noise with high brightness (PNHB) around the target; then, based on this, gradient direction templates are used to weaken background clutter and target-like background clutter around the target; through the fusion of the two methods, noise and clutter around the target can be adaptively filtered out, so that when performing local contrast enhancement in the subsequent process, the potential target region can obtain a higher weight, thus obtaining the final saliency map, improving the accuracy of target detection and possessing superior anti-foreignness capability. Attached Figure Description

[0063] Figure 1This is a flowchart illustrating the infrared weak target detection method based on adaptive contrast enhancement in Embodiment 2 of the present invention.

[0064] Figure 2 The input image in Embodiment 2 of the present invention includes a target and clutter similar to the target.

[0065] Figure 3 These are gradient distribution maps of the target and similar clutter in Embodiment 2 of the present invention, wherein (a) is the gradient distribution map of the target, (b) is the gradient distribution map of clutter 1, (c) is the gradient distribution map of clutter 3, and (d) is the gradient distribution map of clutter 4.

[0066] Figure 4 This is a schematic diagram of histogram weight calculation in Embodiment 2 of the present invention;

[0067] Figure 5 This is a schematic diagram of the fusion result in Embodiment 2 of the present invention, wherein (a) is the input image and its three-dimensional distribution map, (b) is the divergence feature and its three-dimensional distribution map, and (c) is the weighted divergence feature and its three-dimensional distribution map;

[0068] Figure 6 This is a schematic diagram of the detection results in Embodiment 2 of the present invention, wherein (a) is the input image of three weak targets in a complex scene, (b) is the three-dimensional distribution map of the input detection results, and (c) is the detection result of the input image. Detailed Implementation

[0069] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0070] Example 1

[0071] This embodiment provides a technical solution: an infrared weak target detection method based on adaptive contrast enhancement, comprising the following steps:

[0072] Step S1: Read the input infrared image sequence;

[0073] Step S2: Construct a 5×5 local image region with each pixel as the center point, and use manual fitting to fit the local image region into a continuous bivariate cubic function.

[0074] Step S3: Based on the continuous bivariate cubic function obtained in step S2, calculate the gradient magnitude and gradient direction information at the center point to obtain the gradient vector diagram;

[0075] Step S4: Based on the gradient vector map obtained in step S3, calculate the divergence feature corresponding to each pixel as the center point to obtain the divergence feature map.

[0076] Step S5: Based on the gradient direction information obtained in step S2, construct a gradient direction template to calculate the histogram weights at different positions, and obtain the histogram weight map.

[0077] Step S6: After fusing the divergence feature map and histogram weight map obtained in steps S4 and S5 and processing them, a weighted divergence feature map is obtained.

[0078] Step S7: Calculate the local contrast measurement coefficients at each location of the weighted divergence feature map, apply the coefficients at the corresponding locations to the image to obtain the target saliency map, and finally obtain the target location through adaptive threshold segmentation.

[0079] In step S1, the input infrared image I is read. The image size is H×W, where W is the width of the image and H is the height of the image.

[0080] In step S2, a 5×5 local image region is constructed centered on each pixel in the infrared image I, and the local image region is fitted using a manual fitting method. During manual fitting, the infrared local image (local image region) can be fitted into a continuous bivariate cubic function:

[0081]

[0082] Among them, the function g i (r,c) can be obtained from the first-order Chebyshev polynomials R and C by neglecting terms of order four and above, resulting in a symmetric set of second-order Chebyshev polynomials. To replace it, due to the properties of Chebyshev discrete orthogonal polynomials and the least squares method, the coefficients k1,k2,…,k 10 The calculation can be simplified to:

[0083]

[0084] The values ​​of r and c range from [-2, 2] to each point in the 5×5 local image region from left to right. For example, (0, 0) corresponds to the center of the local image, and (-2, -2) corresponds to the top left corner of the local image.

[0085] In step S3, the specific steps for calculating the gradient magnitude and direction at the center point are as follows:

[0086] By manually fitting a local image region, the function f(r,c) for that region is obtained. To solve for the magnitude and direction of the gradient at the center point, the directional derivatives at 0° and 90° corresponding to the center point need to be calculated, denoted as f0.′ and f9 ′ 0:

[0087]

[0088] Where α represents the direction, and the gradient magnitude and direction information corresponding to the center point are represented as follows: and

[0089]

[0090] The range of arctan is: To calculate all directions, we use arctan2 to map the directions in the four quadrants to [-π, π].

[0091] The gradient vector map G can be obtained by calculating the gradient magnitude and direction information at each location in the image.

[0092] In step S4, a 5×5 local image region is constructed in the gradient field with each pixel as the center, and the divergence of the center point is Δf(0,0):

[0093]

[0094] The divergence value of each point is calculated to obtain the divergence feature map D, denoted as D(x,y).

[0095] In step S5, the specific steps for calculating the histogram weight map are as follows:

[0096] In the input image, a 5×5 local image region is constructed centered at each pixel location, and the value at each pixel location is the gradient direction at that point. Since the gradient direction at the target center point frequently changes, the gradient direction value at the center point is removed, leaving matrix M:

[0097] M(r,c)=del(I c )

[0098] Among them, I c This indicates the center position of a 5×5 local image region, and del represents the function to delete a specific element;

[0099] Divide the remaining elements into class b, and calculate the number of elements belonging to each class by comparing them with the gradient direction template, denoted as N(k):

[0100] N(k) = hist(M(k), b)

[0101] Where M(k) represents the matrix region of the k-th class after deleting the gradient direction value at the center point, b represents the number of calculated classes, hist represents the histogram calculation function, and the gradient direction template represents the direction matrix of all positions in the local image region pointing to the center point.

[0102] If the number calculated for the kth category is greater than Will be restricted to Conversely, if it is less than The weight of each category region is determined by the actual number of calculations, and is denoted as W(k):

[0103]

[0104] The weight of the center point is denoted as W(x,y):

[0105]

[0106] Where x and y represent the width and height of a pixel in the original infrared image.

[0107] In step S6, the specific steps for calculating the weighted fused divergence feature map are as follows:

[0108] The divergence feature map and histogram weight map obtained in steps S4 and S5 are fused to obtain the fused divergence feature map, denoted as S(x,y):

[0109] S(x,y)=W(x,y)×D(x,y)

[0110] Since the divergence of the target center region is less than 0, the positive values ​​of the divergence feature map after fusion are ignored. Then, the absolute values ​​of all negative values ​​are taken to obtain the absolute value map of the weighted divergence, represented as A(x,y):

[0111] A(x,y)=|min(0,S(x,y))|

[0112] By weighted fusion of divergence and histogram, noise and clutter around the target region have been filtered out relatively cleanly. To further improve contrast, the absolute value map of the weighted divergence can be squared, resulting in the weighted divergence feature map, denoted as WD(x,y):

[0113] WD(x,y)=A(x,y)×A(x,y)

[0114] In step S7, the specific steps for measuring local contrast are as follows:

[0115] Construct a 5×5 central region, denoted as region 0, using each pixel in the squared-weighted absolute divergence map as the center location. Construct regions of the same size around region 0, denoted as regions 1-8, and calculate the average value for each region.

[0116]

[0117] Where (x,y) represents the element position in the WD diagram, i represents the i-th region, i = 1, 2, ... 8, and j represents the element position in the i-th region.

[0118] The average value of each area surrounding the central area is denoted as m. b :

[0119]

[0120] The contrast measurement coefficients are thus constructed and applied to the WD map to obtain the target saliency map, denoted as Out(x,y):

[0121]

[0122] Where L represents the maximum value in region 0.

[0123] Finally, the target center position can be obtained by using an adaptive threshold T:

[0124] T=μ+k×σ

[0125] Where μ represents the mean of the significance plot, σ represents the standard deviation of the significance plot, and k represents the coefficient, which is generally taken as 2-5.

[0126] Example 2

[0127] like Figure 1 As shown, this embodiment further illustrates the technical solution in Embodiment 1, specifically including the following steps:

[0128] Step 1: Read the input infrared image sequence.

[0129] Step 2: Construct a 5×5 local image region centered on each pixel of the original image. Use a manual fitting method to construct the local image region as a continuous bivariate cubic function. Discretize sets using symmetric second-order Chebyshev polynomials As g i (r,c), using the properties of Chebyshev discrete orthogonal polynomials and the least squares method, the coefficients k1,k2,…,k are... 10 The calculation can be simplified to

[0130] Step 3: Since the gradient field is fundamental for calculating the divergence features and histogram weights of an image, it is necessary to first calculate the magnitude and direction of the gradient at each location in the image. Based on the continuous bivariate cubic function f(r,c) obtained in Step 2, the directional derivative at each location can be expressed as: The directional derivatives at 0° and 90° corresponding to the center point are expressed as f0. ′ and f9 ′ 0. The magnitude of the gradient corresponding to the center point is expressed as: The gradient direction corresponding to the center point is represented as: Due to arctan's To calculate all directions in the four quadrants, we use the arctan2 function to map the direction range to [-π, π]. The gradient vector map G is obtained by calculating the magnitude and direction of the gradient at each location in the image.

[0131] Step 4: Construct a 5×5 local image region centered on each pixel in the gradient field. The divergence of the center point can be expressed as: The divergence value of each point is calculated to obtain the divergence map D, denoted as D(x,y).

[0132] Step 5: As Figure 2 As shown, the input image contains many clutter elements similar to the target, interfering with target detection. For example... Figure 4 As shown, we will use Clutter 3 as an example to specifically introduce the calculation of histogram weights. A 5×5 local image region is constructed with each pixel position as the center, and the value at each position is the gradient direction at that point. Since the gradient direction at the target center point often changes, the gradient direction value at the center point is deleted, and the remaining matrix is ​​M(r,c)=del(I c ), where I c This represents the center position of a 5×5 local image region. The `del` function deletes the element at that position. The remaining elements are divided into class b, and then compared with the gradient direction template (e.g., ...). Figure 4 The comparison (shown on the right) calculates the number of categories each element belongs to, expressed as: N(k) = hist(M(k), b), where M(k) represents the matrix region of the k-th class after removing the gradient direction value at the center point. b represents the number of categories calculated. The hist function represents the histogram calculation. If the number of categories calculated for the k-th class is greater than... Will be restricted to Conversely, if it is less than The weight of each category region is determined by the actual number of calculations, and is expressed as follows: The weight of the center point is denoted as: Where x and y represent the pixel positions in the original image.

[0133] Step 6: Having obtained the divergence and histogram weights of the input image from Steps 4 and 5, they need to be weighted and fused. The fused divergence feature map is represented as: S(x,y)=W(x,y)×D(x,y), as follows. Figure 3 As shown in (b), the 5×5 local image region at the center of the target exhibits a shrinking trend, so the divergence at its center is less than 0. The absolute value of the weighted divergence is represented as: A(x,y)=|min(0,S(x,y))|. Figure 5 As shown in (c), the noise and clutter around the target area have been filtered out relatively cleanly through the weighted fusion of divergence and histogram. In order to further improve the contrast, the absolute value map of the weighted divergence can be squared, denoted as: WD(x,y)=A(x,y)×A(x,y).

[0134] Step 7: As Figure 5 As shown in (c), the target intensity is still lower than the background clutter because the contrast information between the target and the surrounding background has not yet been used. A 5×5 central region is constructed with each pixel in the WD image as the center, denoted as region 0. Regions of the same size as region 0 are constructed around region 0, denoted as regions 1-8, and the average value of each region is calculated: Where (x, y) represents the element position in the WD plot, i represents the i-th region (i = 1, 2, ..., 8), and j represents the element position within the i-th region. The average value of each region surrounding the central region is denoted as... The contrast measurement coefficients are thus constructed and applied to the WD map to obtain the target saliency map, denoted as: Where L represents the maximum value of region 0. Finally, the target center location can be obtained by using an adaptive threshold T: T = μ + k × σ, where μ represents the mean of the saliency map, σ represents the standard deviation of the saliency map, and k represents a coefficient, typically ranging from 2 to 5. The detection results are as follows: Figure 6 As shown.

[0135] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for detecting weak infrared targets based on adaptive contrast enhancement, characterized in that, Includes the following steps: Step S1: Read the input infrared image sequence; Step S2: Construct a 5×5 local image region with each pixel as the center point, and use manual fitting to fit the local image region into a continuous bivariate cubic function. Step S3: Based on the continuous bivariate cubic function obtained in step S2, calculate the gradient magnitude and gradient direction information at the center point to obtain the gradient vector diagram; Step S4: Based on the gradient vector map obtained in step S3, calculate the divergence feature corresponding to each pixel as the center point to obtain the divergence feature map. Step S5: Based on the gradient direction information obtained in step S2, construct a gradient direction template to calculate the histogram weights at different positions, and obtain the histogram weight map. Step S6: After fusing the divergence feature map and histogram weight map obtained in steps S4 and S5 and processing them, a weighted divergence feature map is obtained. Step S7: Calculate the local contrast measurement coefficients at each location of the weighted divergence feature map, apply the coefficients at the corresponding locations to the image to obtain the target saliency map, and finally obtain the target location through adaptive threshold segmentation.

2. The infrared weak target detection method based on adaptive contrast enhancement according to claim 1, characterized in that, In step S1, the input infrared image I is read, and the image size is H×W, where W is the width of the image and H is the height of the image.

3. The infrared weak target detection method based on adaptive contrast enhancement according to claim 2, characterized in that, In step S2, the fitted continuous bivariate cubic function is as follows: Among them, the function g i (r,c) is a set of symmetric second-order Chebyshev polynomials obtained by neglecting terms of fourth order and above from the first-order Chebyshev polynomials R and C. To replace, coefficient k i as follows: Where r and c take values ​​in the range of [-2, 2], corresponding to each point from left to right in the 5×5 local image region.

4. The infrared weak target detection method based on adaptive contrast enhancement according to claim 3, characterized in that, In step S3, the specific process of obtaining the gradient vector diagram is as follows: Step S31: After obtaining the function f(r,c) for the local image region, calculate the directional derivatives at 0° and 90° corresponding to the center point, denoted as f0. ′ and f9 ′ 0: Where α represents the direction; Step S32: Then calculate the gradient magnitude and direction information corresponding to the center point, represented as follows: and The range of arctan is: Step S33: Obtain the gradient vector map G by calculating the gradient magnitude and direction information at each location in the image.

5. The infrared weak target detection method based on adaptive contrast enhancement according to claim 4, characterized in that, In step S4, a 5×5 local image region is constructed in the gradient field with each pixel as the center, and the divergence of the center point is Δf(0,0): The divergence feature map, denoted as D(x,y), can be obtained by calculating the divergence value of each point according to the above formula.

6. The infrared weak target detection method based on adaptive contrast enhancement according to claim 5, characterized in that, In step S5, the specific process of obtaining the histogram weight map is as follows: Step S51: Since the gradient direction at the target center point is time-varying, the gradient direction value at the center point is deleted, and the remaining matrix is ​​M: M(r,c)=del(I c ) Among them, I c This indicates the center position of a 5×5 local image region, and del represents the function to delete a specific element; Step S52: Divide the remaining elements into class b, and calculate the number of elements belonging to the corresponding class by comparing them with the gradient direction template, denoted as N(k): N(k) = hist(M(k), b) Where M(k) represents the matrix region of the k-th class after deleting the gradient direction value at the center point, b represents the number of calculated classes, hist represents the histogram calculation function, and the gradient direction template represents the direction matrix of all positions in the local image region pointing to the center point. Step S53: If the number calculated for the k-th category is greater than... Will be restricted to Conversely, if it is less than The weight of each category region is determined by the actual number of calculations, and is denoted as W(k): Step S54: Calculate the weight of the center point, denoted as W(x,y): Where x and y represent the width and height of a pixel in the original infrared image; Step S55: Calculate the center point weight W(x,y) with each point in the image as the center point to obtain the histogram weight matrix, which is the histogram weight map.

7. The infrared weak target detection method based on adaptive contrast enhancement according to claim 6, characterized in that, In step S6, the specific processing procedure is as follows: Step S61: Merge the divergence feature map and histogram weight map obtained in steps S4 and S5 to obtain the fused divergence feature map, denoted as S(x,y): S(x,y)=W(x,y)×D(x,y); Step S62: Since the divergence of the target center region is less than 0, ignore the positive value part of the fused divergence feature map, and then take the absolute value of all negative values ​​to obtain the absolute value map of the weighted divergence, represented as A(x,y): A(x,y) = |min(0,S(x,y))|; Step S63: Squaring the absolute value map of the weighted divergence yields the squared weighted absolute value map of divergence, i.e., the weighted divergence feature map, denoted as WD(x,y): WD(x,y)=A(x,y)×A(x,y).

8. The infrared weak target detection method based on adaptive contrast enhancement according to claim 7, characterized in that, In step S7, the specific processing procedure is as follows: Step S71: Construct a 5×5 central region, denoted as region 0, with each pixel in the squared weighted absolute divergence map as the center position. Construct regions of the same size around region 0, denoted as regions 1-8. Calculate the average value of each region: Where (x,y) represents the element position of the square-weighted absolute divergence plot, i represents the i-th region, i = 1, 2, ... 8, and j represents the element position in the i-th region; Step S72: Calculate the average value of each area surrounding the central area, denoted as m. b : Based on the average value m b The local contrast measurement coefficients are constructed and applied to the squared-weighted absolute divergence map to obtain the target saliency map, denoted as Out(x,y): Where L represents the maximum value in region 0; Step S73: Obtain the target center position through adaptive threshold T segmentation: T=μ+k×σ Where μ represents the mean of the saliency plot, σ represents the standard deviation of the saliency plot, and k represents the coefficient.

Images