An infrared small target detection method based on information entropy and local contrast weighting

Abstract

The application discloses an infrared small target detection method based on information entropy and local contrast weight, and belongs to the technical field of infrared small target detection, and comprises the following steps: acquiring an original target video binary file; pre-processing a file image; calculating a multi-scale contrast weight and a multi-scale correction entropy; obtaining a multi-scale saliency map according to the multi-scale contrast weight and the multi-scale correction entropy; performing threshold segmentation and binarization processing operation on the multi-scale saliency map, and then detecting a final target position. The application effectively improves the accuracy of small targets in a complex background, further improves the detection efficiency of the algorithm on the basis of existing information entropy detection, and provides a new idea for subsequent information entropy detection algorithm research and engineering application.

Description

Technical Field

[0001] This invention relates to the field of infrared small target detection technology, specifically to an infrared small target detection method based on information entropy and local contrast weighting. Background Technology

[0002] In the field of infrared small target detection in computer imaging, infrared video can provide richer target information than visible light, especially at night or under certain special conditions. Targets with significant thermal radiation, such as birds, aircraft, ships, and fire sources, appear as small, indistinct localized areas under long-distance imaging conditions, making rapid and accurate target detection very challenging. Furthermore, the lack of obvious features and shape information for infrared small targets means that feature-based learning methods cannot be directly used to distinguish them.

[0003] In continuously changing infrared video scenarios, targets are typically constantly shifting, moving from far to near or vice versa, thus exhibiting multi-scale characteristics. The Society of Optoelectronics Engineers (SAE) defines small target sizes as typically ranging from 1×1 to 9×9 pixels based on target imaging. However, existing infrared small target detection methods perform poorly in complex background conditions, exhibiting high false alarm rates and low accuracy. Furthermore, even algorithms with better accuracy suffer from excessively long computation times, failing to meet real-time requirements.

[0004] Chinese invention patent application CN109816641A discloses a method for detecting small infrared targets based on morphological fusion and weighted local information entropy. This method includes three stages: preprocessing, difference calculation, and thresholding. In the preprocessing stage, Top-Hat is used to preprocess the original image at different scales. In the difference calculation stage, the image difference between adjacent scales of Top-Hat is calculated to obtain the minimum difference map, and the local information entropy of the initial image is calculated and fused with the information entropy as a weight to obtain a saliency map. In the thresholding stage, threshold segmentation technology is used to filter the saliency map of the small infrared targets and perform binarization to obtain the location of the small infrared targets. However, the Top-Hat preprocessing method and local information entropy in the above method cannot be well adapted to target detection problems in complex backgrounds, and the detection accuracy is affected by bright background noise. These problems urgently need to be solved. Therefore, this paper proposes a method for detecting small infrared targets based on information entropy and local contrast weighting. Summary of the Invention

[0005] The technical problem to be solved by this invention is: how to effectively improve the accuracy of small targets in complex backgrounds, and provides an infrared small target detection method based on information entropy and local contrast weighting.

[0006] The present invention solves the above-mentioned technical problems through the following technical solution, and the present invention includes the following steps:

[0007] Step S1: Using an uncooled infrared detection camera, record the original target video binary file outdoors in a real scene using the camera's original mode;

[0008] Step S2: Decode the original target video binary file recorded in Step S1, and perform non-uniform correction and data mapping to 8 bits to obtain the original test image;

[0009] Step S3: Based on the obtained test original image, design a scale region that adapts to the size of the target. Assume that the center of the region is the target location and the rest of the region is the background location. Construct a matrix and calculate the Frobenius norm and mean of the matrix containing the target and the background, and remove the target and only contain the background matrix to construct the contrast weight.

[0010] Step S4: Using the overall area of ​​the design, calculate the information entropy of different pixel values ​​in the area where the current position is located, combine it with the contrast weight constructed by the Frobenius norm to obtain the saliency map ratio coefficient of the whole image, and combine it with the pixel matrix of the original image to obtain the saliency map.

[0011] Step S5: Add a central region to the initial design to fit the target size. Repeat steps S3 and S4 to calculate a series of contrast weights and saliency map scaling coefficients. Combine these with the pixel matrix of the original image to obtain multiple saliency maps.

[0012] Step S6: Based on the multiple saliency maps calculated in Step S5, fuse them to solve for the maximum value of all positions in each saliency map to obtain the final saliency map. Calculate the mean and standard deviation of the saliency map, and combine the mean and standard deviation to calculate the segmentation threshold. Perform threshold segmentation and binarization on the saliency map to obtain the saliency map after thresholding and binarization. Find the position with the highest intensity in the saliency map, and then detect the final target position.

[0013] Furthermore, in step S1, using a hardware data transmission device and a software display device connected to the uncooled infrared camera, the camera output mode is set to 14-bit raw image, image size is 640*512, single-channel, single-frame binary file output mode, to record infrared small target motion images with complex backgrounds.

[0014] Furthermore, in step S2, based on the recorded 14-bit infrared small target motion image binary file, a program is written to convert it to 8-bit and use OpenCV library functions to perform non-uniform correction on it, and preprocess the data.

[0015] Furthermore, in step S3, the specific processing procedure is as follows:

[0016] S31: Assume the current coordinates are at row i and column j, denoted as point P, and its pixel value is denoted as P(i,j). The size of the target is a rows and b columns. The a row and b column region centered on P is denoted as region 0. With region 0 as the center, construct a 3a row and 3b column region around it. Each region is a row and b column. The region at the top left corner is denoted as region 1, and the regions in the clockwise direction are denoted as regions 2 to 8 to construct the surrounding regions.

[0017] S32: Construct each region into column vectors and concatenate them together. Calculate the Frobenius norm of the matrix containing the central region and its mean column vector.

[0018] S33: Calculate the Frobenius norm of a matrix without a central region and its mean column vector;

[0019] S34: Calculate the difference between the two Frobenius norms to obtain the contrast weight.

[0020] Furthermore, in step S32, the specific processing procedure is as follows:

[0021] S321: Expand each region column-wise to construct an a×b row, 1 column vector. Concatenate the column vectors of each region according to the region labels to form a new matrix T1 with a×b row, 9 columns.

[0022]

[0023] in, This represents the first value in region 0 constructed by column, where a and b are the row and column numbers of the sub-region, respectively.

[0024] S322: Calculate the mean of T1 row by row to form an a×b row and 1 column vector.

[0025]

[0026]

[0027] Where s represents the number of columns in T1, and t represents the number of rows in T1;

[0028] S323: Transform column vectors Repeatedly expand to an a×b matrix with 9 columns Calculate matrix T1 and Frobenius norm:

[0029]

[0030] Where (i,j) are the coordinates of point P, x,y represent the x-th row and y-th column of matrix T1, and T xy and Represent matrices T1 and T2 respectively. The pixel values ​​in the x-th row and y-th column, F ij Let Frobenius norm be the number of point P.

[0031] Furthermore, in step S33, the specific processing procedure is as follows:

[0032] S331: Remove the column vector from the central region 0 and reconstruct a new matrix B1 with rows of a×b and columns of 8. Then, calculate the mean of matrix B1 by row to form a column vector with rows of a×b and columns of 1.

[0033] S332: Transform column vectors Repeatedly expand to an a*b row and 8 column matrix Calculate matrix B1 and Frobenius norm:

[0034]

[0035] Where (i,j) are the coordinates of point P, and x,y represent the x-th row and y-th column of matrix B1. xy and Represent matrices B1 and B1 respectively. The pixel values ​​in the x-th row and y-th column, It is represented as the Frobenius norm of the decentralized region of point P.

[0036] Furthermore, in step S34, based on the obtained F ij and Calculate the contrast weight D ij :

[0037]

[0038] Where (i,j) are the coordinates of point P, F ij Let P be the Frobenius norm. For the Frobenius norm of the decentralized region, 9 is the F-norm. ij The number of columns in a matrix is ​​8. Number of columns in a matrix.

[0039] Furthermore, in step S4, the specific processing procedure is as follows:

[0040] S41: Based on the new matrix T1 constructed using the calculated Frobenius norm, select the center position P of region 0 as the center position of the current region, with pixel value P(i,j). Calculate the corrected information entropy E. ij :

[0041]

[0042] Where m1 represents the number of pixel value types contained in matrix T1, P s Represents the pixel value of class s, n s The number of pixel values ​​in class s is represented by H, and H and W represent the height and width of the original test image, respectively.

[0043] S42: Contrast weight D calculated from the current pixel. ij and the corrected information entropy E ij The solution is performed once from left to right and top to bottom across the entire image to obtain a contrast weight matrix D and an information entropy matrix E with the same dimensions as the original test image. The significance value at point (i,j) is calculated and represented as follows:

[0044] DE(i,j)=D(i,j)×E(i,j)×P(i,j)

[0045] Where D(i,j) represents the contrast weight at point (i,j), E(i,j) represents the corrected information entropy at point (i,j), and P(i,j) represents the pixel value at point (i,j).

[0046] Furthermore, in step S5, Δh is added outward from the initial center size a×b region to obtain a new center region. Steps S3 and S4 are repeated on this new center region to calculate a new contrast weight. and Obtain the saliency map DE l Continue until the number of pixels in the central region is greater than 80, then obtain l. max A significant graph is denoted as DE. l Where l = 1, ..., l max .

[0047] Furthermore, in step S6, the formulas for calculating the mean and standard deviation of the saliency plot are as follows:

[0048]

[0049]

[0050] Where H and W represent the height and width of the original test image, respectively, DE(i,j) is the significance value of point (i,j), and m is the mean of the saliency map;

[0051] The formula for calculating the segmentation threshold is as follows:

[0052] T = m + k × s

[0053] Where T represents the segmentation threshold, and m and s represent the mean and standard deviation of the saliency map, respectively.

[0054] Compared with the prior art, the present invention has the following advantages: the infrared small target detection method based on information entropy and local contrast weighting effectively improves the accuracy of small targets in complex backgrounds, further improves the detection efficiency of the algorithm on the basis of existing information entropy detection, and provides new ideas for subsequent research and engineering applications of information entropy-based detection algorithms. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the infrared small target detection method based on information entropy and local contrast weighting in Embodiments 1 and 2 of the present invention.

[0056] Figure 2(a) is the first test image in Embodiment 2 of the present invention;

[0057] Figure 2(b) is the second original test image in Embodiment 2 of the present invention;

[0058] Figure 2(c) is the third original test image in Embodiment 2 of the present invention;

[0059] Figure 3 This is a schematic diagram of the initial design of the area labels in Embodiment 2 of the present invention;

[0060] Figure 4 This is a schematic diagram illustrating the method of constructing column vectors in Embodiment 2 of the present invention;

[0061] Figure 5 This is a schematic diagram of the added initial design area in Embodiment 2 of the present invention;

[0062] Figure 6(a) is the detection result of the first test original image in Embodiment 2 of the present invention;

[0063] Figure 6(b) is the detection result of the second test original image in Embodiment 2 of the present invention;

[0064] Figure 6(c) is the detection result of the third test original image in Embodiment 2 of the present invention. Detailed Implementation

[0065] 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.

[0066] Example 1

[0067] like Figure 1 As shown, this embodiment provides a technical solution: an infrared small target detection method based on information entropy and local contrast weighting, comprising the following steps:

[0068] Step S1: Using an uncooled infrared detection camera, record the raw target video binary file outdoors in a real scene using CameraLink in the camera's original mode;

[0069] Step S2: Decode the original target video binary file recorded in Step S1, and perform non-uniform correction and data mapping to 8 bits to obtain the original test image;

[0070] Step S3: Based on the obtained test original image, design a scale region that adapts to the size of the target. Assume that the center of the region is the target location and the rest of the region is the background location. Construct a matrix and calculate the Frobenius norm and mean of the matrix containing the target and the background, and remove the target and only contain the background matrix to construct the contrast weight.

[0071] Step S4: Using the overall area of ​​the design, calculate the information entropy of different pixel values ​​in the area where the current position is located, combine it with the contrast weight constructed by the Frobenius norm to obtain the saliency map ratio coefficient of the whole image, and combine it with the pixel matrix of the original image to obtain the saliency map.

[0072] Step S5: Add a central region to the initial design to fit the target size. Repeat steps S3 and S4 to calculate a series of contrast weights and saliency map scaling coefficients. Combine these with the pixel matrix of the original image to obtain multiple saliency maps.

[0073] Step S6: Based on the saliency map calculated in step S5, the maximum value of all positions in each saliency map is fused to obtain the final saliency map. The mean and standard deviation of the saliency map are calculated, and the segmentation threshold is calculated by combining the mean and standard deviation. The saliency map is then segmented by thresholding and binarized to obtain the saliency map after thresholding and binarization. The position with the highest intensity is found in the saliency map, and the final target position is detected.

[0074] In this embodiment, in step S1, a CameraLink hardware data transmission device and an eBUS Player software display device connected to an uncooled infrared camera are used to set the camera output mode to 14-bit raw image, image size to 640*512, single-channel, single-frame binary file output mode, and record a series of infrared small target motion images with complex backgrounds.

[0075] In this embodiment, in step S2, a program is written to convert the recorded 14-bit infrared small target motion image binary file to 8-bit and use the OpenCV library to perform non-uniform correction and preprocess the data.

[0076] In this embodiment, the specific processing procedure in step S3 is as follows:

[0077] S31: Assuming the target scale is a*b, denoted as region 0, construct 8 a*b regions around the location of a*b as background regions denoted as regions 1 to 8, thus obtaining a 3a*3b region.

[0078] S32: Construct each region into column vectors and concatenate them together. Calculate the Frobenius norm of the matrix containing the central region and its mean column vector.

[0079] S33: Calculate the Frobenius norm of a matrix without a central region and its mean column vector;

[0080] S34: Calculate the difference between the two Frobenius norms to obtain the contrast weight.

[0081] As a further step, design a scale region that adapts to the size of the target:

[0082] Assume the current coordinates are at row i, column j, denoted as point P, and its pixel value is denoted as P(i,j). The target size is a rows and b columns, and the region centered at point P, defined as region 0, is denoted as region a.

[0083] With region 0 as the center, construct a region of 3a rows and 3b columns around it. Each sub-region is a rows and b columns. The region in the upper left corner is denoted as region 1, and the sub-regions in the clockwise direction are denoted as regions 2 to 8 to construct the surrounding regions.

[0084] In this embodiment, the specific process of calculating the Frobenius norm of the matrix containing the central region and its mean column vector in step S32 is as follows:

[0085] S321: Expand each region column-wise to construct an a×b row, 1 column vector. Concatenate the column vectors of each region according to the region labels to form a new matrix T1 with a×b row, 9 columns.

[0086]

[0087] in, This represents the first value in region 0 constructed by column, where a and b are the row and column numbers of the sub-region, respectively.

[0088] S322: Calculate the mean of T1 row by row to form an a×b row and 1 column vector.

[0089]

[0090]

[0091] Where s represents the number of columns in T1, and t represents the number of rows in T1;

[0092] S323: Transform column vectors Repeatedly expand to an a×b matrix with 9 columns Calculate matrix T1 and Frobenius norm:

[0093]

[0094] Where (i,j) are the coordinates of point P, x,y represent the x-th row and y-th column of matrix T1, and T xy and Represent matrices T1 and T2 respectively. The pixel values ​​in the x-th row and y-th column, F ij Let Frobenius norm be the number of point P.

[0095] In this embodiment, the specific process of calculating the Frobenius norm of the matrix without a central region and its mean column vector in step S33 is as follows:

[0096] S331: Remove the column vector from the central region 0 and reconstruct a new matrix B1 with rows of a×b and columns of 8. Then, calculate the mean of matrix B1 by row to form a column vector with rows of a×b and columns of 1.

[0097] S332: Transform column vectors Repeatedly expand to an a*b row and 8 column matrix Calculate matrix B1 and Frobenius norm:

[0098]

[0099] Where (i,j) are the coordinates of point P, and x,y represent the x-th row and y-th column of matrix B1. xy and Represent matrices B1 and B1 respectively. The pixel values ​​in the x-th row and y-th column, It is represented as the Frobenius norm of the decentralized region of point P.

[0100] In this embodiment, in step S34, based on the obtained Fij and Calculate the contrast weight D ij :

[0101]

[0102] Where (i,j) are the coordinates of point P, F ij Let P be the Frobenius norm value. For the Frobenius norm value of the decentralized region, 9 is the F... ij The number of columns in a matrix is ​​8. Number of columns in a matrix.

[0103] In this embodiment, the specific processing procedure in step S4 is as follows:

[0104] S41: Based on the new matrix T1 constructed using the calculated Frobenius norm, select the center position P (row a*b / 2, first column) of region 0 as the center position of the current region, with pixel value P(i,j). Calculate the corrected information entropy E. ij :

[0105]

[0106] Where m1 represents the number of pixel value types contained in matrix T1, P s Represents the pixel value of class s, n s The number of pixel values ​​in class s is represented by H, and H and W represent the height and width of the original test image, respectively.

[0107] S42: Contrast weight D calculated from the current pixel. ij and the corrected information entropy E ij The solution is performed once from left to right and top to bottom across the entire image to obtain a contrast weight matrix D and an information entropy matrix E with the same dimensions as the original test image. The significance value at point (i,j) is calculated and represented as follows:

[0108] DE(i,j)=D(i,j)×E(i,j)×P(i,j)

[0109] Where D(i,j) represents the contrast weight at point (i,j), E(i,j) represents the corrected information entropy at point (i,j), and P(i,j) represents the pixel value at point (i,j).

[0110] In this embodiment, in step S5, Δh is added outward from the initial center size a×b region to obtain a new center region. Steps S3 and S4 are repeated on the new center region to calculate a new contrast weight. and Obtain the saliency map DE l Continue until the number of pixels in the central region is greater than 80, then obtain l. max A significant graph is denoted as DE. l Where l = 1, ..., l max .

[0111] In this embodiment, in step S6, the formulas for calculating the mean and standard deviation of the saliency plot are as follows:

[0112]

[0113] Where H and W represent the height and width of the original test image, respectively, and DE(i,j) is the significance value of point (i,j);

[0114]

[0115] Where H and W represent the height and width of the original test image, respectively, DE(i,j) is the significance value of point (i,j), m is the mean of the saliency map, and s is the standard deviation of the saliency map;

[0116] The formula for calculating the segmentation threshold is as follows:

[0117] T = m + k × s

[0118] Where T represents the segmentation threshold, and m and s represent the mean and standard deviation of the saliency map, respectively.

[0119] Example 2

[0120] like Figure 1 As shown, this embodiment provides an infrared small target detection method based on information entropy and local contrast weighting, including the following steps:

[0121] Step 1: Using the CameraLink hardware data transmission device and eBUS Player software display device connected to the uncooled infrared detection camera, set the camera output mode to 14-bit raw image, image size to 640*512, single-channel, single-frame binary file output mode. Record a series of moving images of small infrared targets with complex backgrounds.

[0122] Step 2: Based on the recorded 14-bit binary infrared small target motion image file, write a Python program to convert it to 8-bit and use the OpenCV library to perform non-uniformity correction and data preprocessing. For example... Figures 2(a) to 2(c) As shown, this is the original test image obtained for processing by the detection method.

[0123] Step 3: As Figure 3As shown, assuming the target is located at the center of region 0, with a size of a×b, eight surrounding regions of a×b are created around region 0 in the obtained test image, labeled 1-8. For example... Figure 4 As shown, taking region 0 as an example, we construct an a×b row and 1 column vector by expanding the columns, and then concatenate the column vectors of each region according to the region labels to form a new matrix of a×b row and 9 columns. Then calculate the mean of matrix T1 row by row to form an a×b row, 1 column vector. column vector Repeatedly expand to an a×b matrix with 9 columns Calculate matrix T1 and Frobenius norm Then, remove the column vector from the center region 0 and reconstruct a new matrix B1 with rows of a×b and columns of 8. Calculate the mean of matrix B1 row by row to form a column vector with rows of a×b and columns of 1. column vector Repeatedly expand to an a×b matrix with 8 columns Calculate matrix B1 and Frobenius norm Based on the obtained Frobenius norm F ij and Calculate contrast weight

[0124] Step 4: Based on the new matrix T1 constructed using the calculated Frobenius norm, select the center position of region 0 (the first column of the a×b / 2th row) as the center position of the current region, with its pixel value P(i,j). Calculate the corrected information entropy E. ij :

[0125] The contrast weight D calculated from the current pixel ij and the corrected information entropy E ij The contrast weight matrix D and information entropy matrix E are obtained by solving the problem from left to right and from top to bottom across the entire image. The significance value at point (i,j) can be calculated as: DE(i,j)=D(i,j)×E(i,j)×P(i,j). The significance map denoted as DE1 can be obtained from all the significance values.

[0126] Step 5: As Figure 5 As shown, a new central region is obtained by increasing Δh outward from the initial central region of size a×b. Steps 3 and 4 are repeated on this new central region to calculate the new contrast weight. and corrected information entropy Obtain the saliency map DE lContinue until the number of pixels in the central region is greater than 80, then obtain l. max A salient map DE l Where l = 1, ..., l max .

[0127] Step 6: Based on the l calculated in Step 5 max A salient map DE l The final saliency map is obtained by fusing and solving for the maximum value at all positions in each saliency map, and then the mean of this saliency map is calculated. and standard deviation The segmentation threshold T = m + k × s is calculated by combining the mean and standard deviation. Thresholding and binarization are then performed on the saliency map to obtain a thresholded and binarized saliency map. The location (i, j) with the highest intensity is identified as the center of the detected target. The size of the target, a + l * Δh and b + l * Δh, can be calculated from the l-th saliency map containing the maximum intensity value. This allows for accurate target location detection. The detection results are as follows: Figures 6(a) to 6(c) As shown.

[0128] The following is a specific example of obtaining the final saliency map by fusing the maximum value at all positions of each saliency map: Assume l max The value is 3, meaning there are three saliency maps DE1, DE2, and DE3. These three saliency map matrices have the same length and width. For each position in the matrix, the maximum value among the three matrices is used as the saliency value of the new matrix, thus obtaining the final saliency map.

[0129] In summary, the infrared small target detection method based on information entropy and local contrast weighting described above effectively improves the accuracy of small targets in complex backgrounds. It further enhances the detection efficiency of the algorithm based on existing information entropy detection methods, and provides new ideas for subsequent research and engineering applications of information entropy-based detection algorithms.

[0130] 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 small infrared targets based on information entropy and local contrast weighting, characterized in that, Includes the following steps: Step S1: Using an uncooled infrared detection camera, record the original target video binary file outdoors in a real scene using the camera's original mode; Step S2: Decode the original target video binary file recorded in Step S1, and perform non-uniform correction and data mapping to 8 bits to obtain the original test image; Step S3: Based on the obtained test original image, design a scale region that adapts to the size of the target. Assume that the center of the region is the target location and the rest of the region is the background location. Construct a matrix and calculate the Frobenius norm and mean of the matrix containing the target and the background, and remove the target and only contain the background matrix to construct the contrast weight. Step S4: Using the overall area of ​​the design, calculate the information entropy of different pixel values ​​in the area where the current position is located, combine it with the contrast weight constructed by the Frobenius norm to obtain the saliency map ratio coefficient of the whole image, and combine it with the pixel matrix of the original image to obtain the saliency map. Step S5: Add a central region to the initial design to fit the target size. Repeat steps S3 and S4 to calculate a series of contrast weights and saliency map scaling coefficients. Combine these with the pixel matrix of the original image to obtain multiple saliency maps. Step S6: Based on the multiple saliency maps calculated in Step S5, fuse them to solve for the maximum value of all positions in each saliency map to obtain the final saliency map. Calculate the mean and standard deviation of the saliency map, and combine the mean and standard deviation to calculate the segmentation threshold. Perform threshold segmentation and binarization on the saliency map to obtain the saliency map after thresholding and binarization. Find the position with the highest intensity in the saliency map, and then detect the final target position.

2. The infrared small target detection method based on information entropy and local contrast weighting according to claim 1, characterized in that, In step S1, using a hardware data transmission device and a software display device connected to an uncooled infrared camera, the camera output mode is set to 14-bit raw image, image size is 640*512, single-channel, single-frame binary file output mode, to record an infrared small target motion image with a complex background.

3. The infrared small target detection method based on information entropy and local contrast weighting according to claim 1, characterized in that, In step S2, based on the recorded 14-bit infrared small target motion image binary file, a program is written to convert it to 8-bit and use OpenCV library functions to perform non-uniform correction and preprocess the data.

4. The infrared small target detection method based on information entropy and local contrast weighting according to claim 1, characterized in that, In step S3, the specific processing procedure is as follows: S3l: Assume the current coordinates are at the i-th row and j-th column, denoted as point P, and its pixel value is denoted as P(i,j). The size of the target is a rows and b columns. The a-row and b-column region centered on P is denoted as region 0. With region 0 as the center, construct a 3a-row and 3b-column region around it. Each region is a rows and b columns. The region at the top left corner is denoted as region 1, and the regions in the clockwise direction are denoted as regions 2 to 8 to construct the surrounding regions. S32: Construct each region into column vectors and concatenate them together. Calculate the Frobenius norm of the matrix containing the central region and its mean column vector. S33: Calculate the Frobenius norm of a matrix without a central region and its mean column vector; S34: Calculate the difference between the two Frobenius norms to obtain the contrast weight.

5. The infrared small target detection method based on information entropy and local contrast weighting according to claim 4, characterized in that, In step S32, the specific processing procedure is as follows: S321: Expand each region column-wise to construct an a×b row, 1 column vector. Concatenate the column vectors of each region according to the region labels to form a new matrix T1 with a×b row, 9 columns. in, This represents the first value in region 0 constructed by column, where a and b are the row and column numbers of the sub-region, respectively. S322: Calculate the mean of T1 row by row to form an a×b row and 1 column vector. Where s represents the number of columns in T1, and t represents the number of rows in T1; S323: Transform column vectors Repeatedly expand to an a×b matrix with 9 columns Calculate matrix T1 and Frobenius norm: Where (i, j) are the coordinates of point P, x, y represent the x-th row and y-th column of matrix T1, T xy and Represent matrices T1 and T2 respectively. The pixel values ​​in the x-th row and y-th column, F ij Let Frobenius norm be the number of point P.

6. The infrared small target detection method based on information entropy and local contrast weighting according to claim 5, characterized in that, In step S33, the specific processing procedure is as follows: S331: Remove the column vector from the central region 0 and reconstruct a new matrix B1 with rows of a×b and columns of 8. Then, calculate the mean of matrix B1 by row to form a column vector with rows of a×b and columns of 1. S332: Transform column vectors Repeatedly expand to an a*b row and 8 column matrix Calculate matrix B1 and Frobenius norm: Where (i, j) are the coordinates of point P, x, y represent the x-th row and y-th column of matrix B1, B xy and Represent matrices B1 and B1 respectively. The pixel values ​​in the x-th row and y-th column, It is represented as the Frobenius norm of the decentralized region of point P.

7. The infrared small target detection method based on information entropy and local contrast weighting according to claim 6, characterized in that, In step S34, based on the obtained F ij and Calculate the contrast weight D ij : Where (i, j) are the coordinates of point P, F ij Let P be the Frobenius norm. For the Frobenius norm of the decentralized region, 9 is the F-norm. ij The number of columns in a matrix is ​​8. Number of columns in a matrix.

8. The infrared small target detection method based on information entropy and local contrast weighting according to claim 7, characterized in that, In step S4, the specific processing procedure is as follows: S41: Based on the new matrix T1 constructed using the calculated Frobenius norm, select the center position P of region 0 as the center position of the current region, with pixel value P(i,j). Calculate the corrected information entropy E. ij : Where m1 represents the number of pixel value types contained in matrix T1, P s Represents the pixel value of class s, n s The number of pixel values ​​in class s is represented by H, and H and W represent the height and width of the original test image, respectively. S42: Contrast weight D calculated from the current pixel. ij and the corrected information entropy E ij The solution is performed once across the entire image from left to right and top to bottom to obtain a contrast weight matrix D and an information entropy matrix E with the same dimensions as the original test image. The significance value at point (i, j) is calculated and represented as follows: DE(i,j)=D(i,j)×E(i,j)×P(i,j) Where D(i,j) represents the contrast weight at point (i,j), E(i,j) represents the corrected information entropy at point (i,j), and P(i,j) represents the pixel value at point (i,j).

9. The infrared small target detection method based on information entropy and local contrast weighting according to claim 8, characterized in that, In step S5, Δh is added outward from the initial center size a×b region to obtain a new center region. Steps S3 and S4 are repeated on the new center region to calculate a new contrast weight. and Obtain the saliency map DE l Continue until the number of pixels in the central region is greater than 80, then obtain l. max A significant graph is denoted as DE. l , where l = 1, ..., l max .

10. The infrared small target detection method based on information entropy and local contrast weighting according to claim 9, characterized in that, In step S6, the formulas for calculating the mean and standard deviation of the saliency plot are as follows: Where H and W represent the height and width of the original test image, respectively, DE(i,j) is the significance value of point (i,j), and m is the mean of the saliency map; The formula for calculating the segmentation threshold is as follows: T = m + k × s Where T represents the segmentation threshold, and m and s represent the mean and standard deviation of the saliency map, respectively.

Images