Infrared target detection method based on entropy and distance weighted features in complex background

Abstract

The application discloses an infrared target detection method based on entropy and distance weighted features in a complex background and relates to the technical field of infrared image target detection.The application comprises the following steps: adopting an improved firefly algorithm to perform threshold segmentation on an infrared image; adopting a morphological algorithm to perform noise reduction and repair on the segmented image; adopting a connected region based on geometric parameter constraint to mark the image to obtain a region of interest; performing distance weighted feature extraction on the region of interest to obtain a first feature vector; performing entropy weighted feature extraction on the region of interest to obtain a second feature vector; weighting and fusing the first feature vector and the second feature vector to obtain a fused feature vector; and inputting the fused feature vector into a support vector machine for training and classification to obtain a target detection result.The application realizes a simpler calculation mode, optimizes the complexity of the whole algorithm, effectively saves operation resources and is beneficial to hardware implementation of the algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infrared image target detection, and more particularly to an infrared target detection method based on entropy and distance weighted features in a complex background. BACKGROUND

[0002] With the development of target detection technology, infrared target detection technology has been widely studied and applied in precision guidance, night vision and night warfare, and long-range early warning. Infrared imaging has the advantages of strong concealment, high sensitivity, strong smoke penetration ability, and high detection accuracy. Compared with visible light imaging, the infrared imaging system can detect a longer distance. However, due to the limitation of the principle of infrared imaging, the obtained infrared image is a gray image, and the detailed information of the infrared target is less, the contrast of the whole image is low, and the signal-to-noise ratio of the infrared image is low with the increase of the detection distance. At present, the existing infrared image target detection algorithm has a high false alarm rate for complex background and low signal-to-noise ratio image target detection, and the real-time performance of the target detection algorithm is poor, which is difficult to meet the needs of engineering application.

[0003] Chinese patent publication No. "CN113822279A" entitled "Infrared target detection method, device, equipment and medium based on multi-feature fusion" first uses global threshold segmentation to obtain a binary image. Then, the binary image is labeled by connected domain to obtain a candidate target, and the feature components of the candidate target are extracted. Secondly, the multi-feature fusion of the candidate target is carried out to obtain a normalized feature vector. Finally, it is judged whether the normalized feature vector reaches the detection threshold. If the detection threshold is reached, the candidate target is retained, otherwise the candidate target is deleted. The infrared image target detection method has low detection accuracy for complex background infrared images, and the whole algorithm has high complexity, and cannot realize real-time detection.

[0004] In order to solve the above problems, the present application provides an infrared target detection method based on entropy and distance weighted features in a complex background to improve the detection efficiency of infrared image target detection, which is a problem urgently needed to be solved by those skilled in the art. SUMMARY

[0005] Therefore, the present application provides an infrared target detection method based on entropy and distance weighted features in a complex background, which solves the problem of low detection accuracy of the existing infrared image target detection method in a complex background, and achieves the purpose of improving the detection efficiency of infrared image target detection.

[0006] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0007] The infrared target detection method based on entropy and distance weighted features in a complex background comprises the following specific steps:

[0008] S1, the improved firefly algorithm is used for threshold segmentation of the infrared image;

[0009] S2, a morphological algorithm is used for noise reduction and repair on the segmented image to obtain a preprocessed image;

[0010] S3, a connected region based on geometric parameter constraint is used for marking the preprocessed image to obtain a region of interest;

[0011] S4, distance-weighted feature extraction is performed on the region of interest to obtain a first feature vector;

[0012] S5, entropy-weighted feature extraction is performed on the region of interest to obtain a second feature vector;

[0013] S6, the first feature vector and the second feature vector are weighted and fused to obtain a fused feature vector;

[0014] S7, the fused feature vector is input into a support vector machine for training and classification to obtain a target detection result.

[0015] Optionally, the specific steps of S1 include:

[0016] S1.1, parameters of the firefly algorithm are set;

[0017] S1.2, the brightness corresponding to each firefly is calculated;

[0018] S1.3, the position of the firefly is updated according to a firefly movement formula model;

[0019] S1.4, the firefly with the maximum brightness is found and recorded;

[0020] S1.5, when the maximum number of iterations is reached and the movement distance of the firefly is less than a preset movement distance threshold, the position of the firefly with the maximum brightness is recorded as (s, t) to obtain a position array; otherwise, S1.2-S1.5 are repeated; wherein the position array is an array obtained by the best threshold value of the two-dimensional image threshold segmentation method in this round of iteration.

[0021] Optionally, the specific steps of S1.1 include:

[0022] a firefly population is initialized;

[0023] the maximum number of iterations is set as M;

[0024] the number of fireflies in the population is set as N; wherein M and N are integers greater than or equal to 1;

[0025] a random step factor is set as α;

[0026] the light intensity absorption coefficient of the propagation medium is set as γ;

[0027] set the initial attraction degree as β;

[0028] initialize the position of the glowworm.

[0029] Optionally, the specific steps of S2 include:

[0030] adopting open operation to remove noise from the foreground region in the infrared image;

[0031] adopting closed operation to repair the foreground region in the infrared image.

[0032] Optionally, the specific steps of S3 include:

[0033] limiting the shape factor of the connected region to narrow the range of the connected region;

[0034] adopting 8-neighborhood method to determine whether a set of pixels belongs to a connected component;

[0035] determining the set of pixels belonging to the connected component as the final region of interest.

[0036] Optionally, the specific steps of S4 include:

[0037] classifying the 8 pixel points in the neighborhood of each region of interest according to distance, and calculating the local binary pattern value of the center pixel point;

[0038] calculating the corresponding weight value of the calculated local binary pattern value;

[0039] calculating the final binary pattern value of the center pixel point according to the weight value;

[0040] calculating the histogram of each region of interest, and performing normalization processing on the histogram;

[0041] concatenating each normalized histogram to obtain the first feature vector of the region of interest.

[0042] Optionally, the specific steps of S5 include:

[0043] performing normalization processing on the region of interest;

[0044] calculating the direction and amplitude of the gradient of all pixel points in the image of the region of interest;

[0045] dividing the image into A cell units, and calculating the gradient histogram of each cell unit; wherein A is a positive integer greater than 1;

[0046] determining the key cell unit and the magnification coefficient;

[0047] Recombine B cell units in A cell units to obtain a block, and all blocks constitute an image same as the original image; wherein B is a positive integer greater than 1, and the image constituted by the blocks is same as the original image;

[0048] Serially connect the cell unit features in each block to obtain the features of the block;

[0049] Serially connect the features of all blocks in the image to obtain the final second feature vector of the image. Optionally, the specific steps of S6 include:

[0050] Normalizing the first feature vector and the second feature vector obtained in S4 and S5;

[0051] Weighted fusion of the normalized first feature vector and the normalized second feature vector;

[0052] Obtaining the final fusion feature vector.

[0053] Optionally, the specific steps of S7 include:

[0054] Inputting the fusion feature vector obtained in S6 into a support vector machine for classification training;

[0055] Obtaining a classifier model after C times of training of the support vector machine, wherein C is a positive integer greater than 1;

[0056] Classifying the extracted feature vector using the classifier model to obtain the final target detection result.

[0057] According to the technical solution, compared with the prior art, the present disclosure provides an infrared target detection method based on entropy and distance weighted features in a complex background, so that the following beneficial effects can be obtained:

[0058] 1. The present disclosure improves the threshold segmentation of the infrared image through the improved firefly algorithm, realizes the accurate segmentation and positioning of the infrared image target in the complex background, improves the target detection probability, and reduces the false alarm rate, aiming at the problems of low contrast and less target feature information of the infrared image in the complex background.

[0059] 2. The present disclosure improves the robustness and stability of the detection result by using the entropy and distance weighted feature extraction method and then weightedly fusing the two features, aiming at the problems of insufficient feature extraction ability and weak robustness of the existing target detection algorithm.

[0060] 3. The infrared target detection algorithm proposed in the present disclosure adopts a more simple calculation method, optimizes the complexity of the entire algorithm, effectively saves the operation resources, and is conducive to the hardware implementation of the algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute a part of the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of the provided drawings.

[0062] Figure 1 The flow chart of the infrared target detection algorithm of the present application;

[0063] Figure 2 The connected region marking result chart under the first complex scene of the present application;

[0064] Figure 3 The connected region marking result chart under the second complex scene of the present application;

[0065] Figure 4 The target detection result chart under the third complex background of the present application;

[0066] Figure 5 The target detection result chart under the fourth complex background of the present application. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0068] As shown in Figure 1 The embodiments of the present application disclose an infrared target detection method based on entropy and distance weighted features under complex background, and the specific steps include:

[0069] S1, performing threshold segmentation on the infrared image by using an improved firefly algorithm;

[0070] S2, performing noise reduction and repair on the segmented image by using a morphological algorithm to obtain a preprocessed image;

[0071] S3, performing labeling on the preprocessed image by using a connected region based on geometric parameter constraint to obtain a region of interest;

[0072] S4, performing feature extraction on the region of interest based on distance weighting to obtain a first feature vector;

[0073] S5, performing feature extraction on the region of interest based on entropy weighting to obtain a second feature vector;

[0074] S6, the first feature vector, the second feature vector is weighted fusion, get fusion feature vector;

[0075] S7, the fusion feature vector is input into support vector machine training and classification, get target detection result. The specific experimental results are shown in Figure 4 、 Figure 5

[0076] Further, the specific steps of S1 include:

[0077] S1.1, set the parameters of firefly algorithm;

[0078] S1.2, calculate the brightness corresponding to each firefly;

[0079] S1.3, according to the firefly movement formula model, update the position of firefly;

[0080] S1.4, find and record the brightest firefly;

[0081] S1.5, when the maximum iteration number is reached and the firefly moving distance is less than the preset moving distance threshold, the position of the brightest firefly is recorded as (s, t), and the position array is obtained; otherwise, repeat S1.2-S1.5; wherein, the position array is the array obtained by the best threshold value of this round of iteration two-dimensional image threshold segmentation method.

[0082] Further, in S1.2, the brightness of firefly is calculated, that is, the maximum between-class variance of two-dimensional image is calculated, and the specific calculation process is as follows:

[0083] Let f(x,y) be the gray value of the position (x,y) of the image I M×N , and the gray level is L, then f(x,y)∈[0,L-1];

[0084] If the total number of pixels of the gray level is f i , then the probability of the i-th gray level is:

[0085]

[0086] In the formula: i represents the gray level, f i represents the number of pixels, P(i) represents the probability of the i-th gray level, M represents the height of the image, and N represents the width of the image; wherein, i=0,1,…,L-1, and

[0087] ​The pixels in the image are divided into two categories based on their gray levels using a threshold t: background C0 and target C1. The gray level of background C0 is 0 to t-1, and the gray level of target C1 is t to L-1. The pixels corresponding to background C0 and target C1 are {f(x,y)<t} and {f(x,y)≥t}, respectively.

[0088] The probability of background C0 occurring is:

[0089]

[0090] In the formula: ω0 represents the probability of the background C0 part appearing;

[0091] The probability of the target C1 part appearing is:

[0092]

[0093] In the formula: ω1 represents the probability of the target C1 part appearing; where ω0+ω1=1;

[0094] The average grayscale value of the background C0 region is:

[0095]

[0096] In the formula: μ0 represents the average gray value of the background C0 part, and t represents the gray level classification threshold;

[0097] The average grayscale value of the target C1 part is:

[0098]

[0099] In the formula: μ1 represents the average gray value of the target C1 part;

[0100] The overall average grayscale value of the image is:

[0101]

[0102] In the formula: μ represents the total average gray value of the image;

[0103] The inter-class variances of the background and target in the image are:

[0104] δ 2 (k)=ω0(μ-μ0) 2 +ω1(μ-μ1) 2

[0105] Where: δ 2 (k) Inter-class variance of background and target under different k values,

[0106] Let the value of k vary from 0 to L-1, and calculate the inter-class variance δ for different values ​​of k. 2 (k) such that δ2 (k) the k value at the maximum is the optimal threshold required, and the threshold size obtained by the algorithm is between them.

[0107] When the maximum number of iterations is reached and the glowworm moves a distance small enough, i.e. |X' i -X i | < 10 -2 , record the position (s, t) of the glowworm with the maximum brightness, and segment the grayscale image with the threshold value.

[0108] Further, the specific steps of S1.1 include:

[0109] Initialize the glowworm population;

[0110] Set the maximum number of iterations M;

[0111] Set the number of population glowworms N; wherein M, N are integers greater than or equal to 1;

[0112] Set the random step factor to be α;

[0113] Set the light intensity absorption coefficient of the propagation medium to be γ;

[0114] Set the initial attraction degree to be β;

[0115] Initialize the position of the glowworm.

[0116] Further, the specific steps of S2 include:

[0117] Use the open operation to remove noise from the foreground region in the infrared image;

[0118] Use the closed operation to repair the foreground region in the infrared image.

[0119] Further, in S2, the open operation is an erosion operation followed by a dilation operation, which makes the outline of the target object smooth. Since the erosion operation is performed first, the open operation can remove objects smaller than the structure element, disconnect narrow connecting parts, and smooth out protrusions on the edges of the target object.

[0120] The closed operation is a dilation operation followed by an erosion operation, which also makes the outline of the target object smooth. Since the dilation operation is performed first, the closed operation can fill in holes in the binary image that are smaller than the structure element, and it can also fill in narrow broken bands contained in the outline of the object.

[0121] A 3x3 matrix is selected as the structure element for morphological operation. The binary image after morphological processing eliminates noise and burrs, reduces some interference contained in the background, and highlights the target region in the foreground region. The calculation formulas for the open operation and the closed operation are as follows:

[0122]

[0123]

[0124] I ks (i,j) represents the image filled with holes, I kc (i,j) represents the image after the closing operation, I ko (i,j) represents the image after the opening operation.

[0125] Further, the specific steps of S3 include:

[0126] The shape factor of the connected region is limited, and the range of the connected region is reduced;

[0127] The method of 8-neighborhood is used to determine whether a set of pixels belongs to a connected component;

[0128] The set of pixels belonging to the connected component is determined as the final region of interest.

[0129] Further, in S3, for the connected region, the sum of the number of all pixel points in its region represents the area size of the connected region, denoted by S, as shown in the following formula:

[0130] S = {the number of all pixel points in the connected region}

[0131] Each connected domain is marked by a regular quadrilateral, and each marked quadrilateral has a corresponding length and width; then the aspect ratio T can be represented as:

[0132]

[0133] In the formula, H represents the height of the quadrilateral, and W represents the width of the quadrilateral; the value range of T is (0, +∞), when T is less than 1, the shape of the connected domain is close to flat, and when T is greater than 1, the shape of the connected domain is elongated.

[0134] The shape factor considers the relationship between the perimeter and the area of the circumscribed rectangle of the connected region, denoted by s:

[0135]

[0136] In the formula, s represents the shape factor, B represents the perimeter of the circumscribed rectangle of the connected region, and A represents the area of the circumscribed rectangle of the connected region.

[0137] The shape factor can describe the shape information of the connected domain; the value range of the shape factor s is (1, +∞).

[0138] The target detected by the present application needs to meet the following conditions, wherein P p , T p , s p are the area, the aspect ratio and the shape factor of the connected domain of the suspected pedestrian target respectively;

[0139] P p / 100≤α

[0140]

[0141] σ1≤s p ≤σ2

[0142] Wherein, alpha represents the threshold value of the area parameter of the suspected pedestrian target connected domain, mu1 represents the threshold value of the minimum aspect ratio parameter of the suspected pedestrian target connected domain, mu2 represents the threshold value of the maximum aspect ratio parameter of the suspected pedestrian target connected domain, sigma1 represents the threshold value of the minimum shape factor parameter of the suspected pedestrian target connected domain, and sigma2 represents the threshold value of the maximum shape factor parameter of the suspected pedestrian target connected domain; wherein, alpha=5, the aspect ratio mu1=1.8, mu2=4.8, the minimum threshold value of the shape factor is sigma1=1.38, and the maximum threshold value of the shape factor is sigma2=1.92.

[0143] Further, the specific steps of the S4 include:

[0144] Classify the 8 pixel points in the neighborhood of each region of interest according to the distance, and calculate the local binary pattern value and of the center pixel point;

[0145] Calculate the corresponding weight value of the calculated local binary pattern value;

[0146] According to the weight value, calculate the final center pixel point binary pattern value;

[0147] Calculate the histogram of each region of interest, and normalize the histogram;

[0148] Concatenate each normalized histogram to obtain the first feature vector of the region of interest.

[0149] Further, in S4, the distance-weighted feature extraction method first classifies the neighborhood pixel points according to the distance, and then combines the code words of two distances to form a new code word;

[0150] Let (x c , y c ) represent the center pixel point, i c represent the gray value of the center pixel point, and i k(k = 1, 2, 3, 4) represent the gray value of the neighborhood pixel point, the gray value of the neighborhood pixel point and the intermediate pixel point is locally binary pattern coded, the rule is:

[0151]

[0152] In the formula: S k represents the gray value of the local binary pattern coding; wherein,

[0153] Because the distance value of the field pixel point and the center pixel point is 1 and two, according to the above coding mode, two binary codes can be coded, that is, c1c2c3c4c5c6c7c8 and c'1c'2c'3c'4c'5c'6c'7c'8, which is converted to decimal, the local binary pattern value of the center pixel point is obtained, and this value is used to represent the gray scale of the region around the center pixel point:

[0154]

[0155] In the formula: L represents the gray scale of the region around the center pixel point;

[0156] The local binary pattern value of the center pixel point obtained after calculation is marked as m1 and m2, wherein m1 corresponds to the neighborhood coding with a distance of 1, and m2 corresponds to the neighborhood coding with a distance of ;

[0157] Then, for the calculated m1 and m2, the corresponding weight needs to be determined. Because the distance value of the neighborhood and the center pixel point is 1 and the closer the neighborhood pixel point to the center point, the greater the correlation coefficient between the neighborhood pixel point coding and the center pixel point, so the weight of the two different neighborhood pixel points is redistributed, and the specific allocation is:

[0158]

[0159] In the formula: l represents the weight of the neighborhood coding with a distance of 1, and l' represents the weight of the neighborhood coding with a distance of 1.

[0160] Wherein, l and l' satisfy l + l' = 1.

[0161] Finally, after determining the weight, the local binary pattern value of the center pixel point is calculated as follows

[0162]

[0163] In the formula: represents the final local binary pattern value of the center pixel point, and [] represents rounding.

[0164] Further, the specific steps of S5 include:

[0165] normalizing the region of interest;

[0166] calculating the direction and magnitude of the gradient of all pixel points in the image of the region of interest;

[0167] dividing the image into A cell units and calculating the gradient histogram of each cell unit; wherein A is a positive integer greater than 1;

[0168] determining the key cell unit and the magnification coefficient;

[0169] recombining B cell units in the A cell units to obtain a block, and all blocks constitute an image identical to the original image; wherein B is a positive integer greater than 1, and the image composed of blocks is identical to the original image;

[0170] concatenating the cell unit features in each block to obtain the features of the block;

[0171] concatenating the features of all blocks in the image to obtain the final second feature vector of the image. Further, in S5, the local region size is selected as one block, and one block is composed of 4 cell units. Since the entropy value is solved in each cell unit, too many segmentation intervals will cause the local region to be very sensitive to small attitude changes. Therefore, the block is selected as the entropy value change unit;

[0172] The entropy value of each block is solved as follows:

[0173]

[0174]

[0175] In the formula: P b represents the entropy value of each block, f(i,j) represents the probability of the (i,j) luminance value, P ij represents the entropy value of (i,j) luminance value; wherein after obtaining the feature of each block of the region of interest, the corresponding entropy value is weighted as a weight to obtain an entropy-weighted feature vector.

[0176] Further, the specific steps of S6 include:

[0177] normalizing the first feature vector and the second feature vector obtained in S4 and S5;

[0178] weighting and fusing the normalized first feature vector and the normalized second feature vector;

[0179] obtaining the final fused feature vector.

[0180] Further, the fusion feature vector obtained in S6 is input into the support vector machine for classification training;

[0181] The support vector machine obtains a classifier model after C times of training, wherein C is a positive integer greater than 1;

[0182] The extracted feature vector is classified using the classifier model to obtain a final target detection result.

[0183] Further, the specific steps of S7 include:

[0184] The fusion feature vector obtained in S6 is input into the support vector machine for classification training;

[0185] The support vector machine obtains a classifier model after multiple times of training, wherein the multiple times of training is an integer greater than 1;

[0186] The extracted feature vector is classified using the classifier model to obtain a final target detection result.

[0187] Further, in S7, the core idea of the support vector machine is to find a hyperplane that can separate different class samples in the sample space, and in a two-dimensional space, it is to find a straight line.

[0188] For a given sample set

[0189] {x i ,y i}, i = 1, 2,..., n

[0190] wherein x i represents a sample, y i ∈{-1, 1} represents a class label of the sample, and n represents the number of samples.

[0191] The support vector is located on the surface that maximizes the interval between the two classes; the support vector is located on the planes wx+b=1 and wx+b=-1, and multiplying the class label y in front of it obtains the expression of the support vector y(wx+b)=1.

[0192] According to the knowledge of analytic geometry, the interval distance is To maximize the interval distance is to minimize The optimization problem can be expressed as follows:

[0193]

[0194] wherein y represents a class number, w represents a slope of the support vector, and b represents an offset of the support vector;

[0195] The solution of the optimization problem can be solved by Lagrange multiplier method, and the following Lagrange objective function is established:

[0196]

[0197] In the formula, L(w, b, a) represents the Lagrange objective function, and a is the Lagrange multiplier;

[0198] The partial derivatives of w and b are solved respectively, and the partial derivatives are zero, and the following formula is obtained:

[0199]

[0200] The above formula is brought into the Lagrange objective function, combined with the constraint condition, and according to the dual theory, the final optimization formula is obtained:

[0201]

[0202] In the formula, max α represents the maximum value of the Lagrange multiplier, x j represents another group of sample points, y j represents the class number corresponding to another group of sample points;

[0203] According to the training sample, the support vector of the sample and the approximate solution w' and b' of the hyperplane parameters w and b can be solved by optimization. At this time, the to-be-measured sample vector x can be judged.

[0204] The classification method of the support vector machine does not need a large amount of training sample data, and the final classification function of the support vector machine is determined by a small number of support vectors, and the support vector also determines the complexity of the calculation, so whether the calculation of the support vector machine is complex or not is irrelevant to the dimension of the feature vector, and the problem caused by the high dimension is avoided to a certain extent. Therefore, the method of the support vector machine is adopted to classify the target features extracted from the infrared image.

[0205] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same and similar parts between each embodiment can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the related parts can be referred to the method part.

[0206] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the appended claims are intended to cover all such modifications that do not depart from the true spirit and scope of the application. Therefore, the application is not limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An infrared target detection method based on entropy and distance-weighted features under complex backgrounds, characterized in that, The specific steps include: S1. Threshold segmentation of infrared images is performed using an improved firefly algorithm; S2. Morphological algorithms are used to denoise and repair the segmented image to obtain a preprocessed image; S3. The preprocessed image is marked using connected regions based on geometric parameter constraints to obtain the region of interest; S4. Perform distance-weighted feature extraction on the region of interest to obtain the first feature vector; S5. Perform entropy-weighted feature extraction on the region of interest to obtain the second feature vector; S6. Weightedly fuse the first feature vector and the second feature vector to obtain the fused feature vector; S7. Input the fused feature vector into the support vector machine for training and classification to obtain the target detection results; The specific steps of S1 include: S1.1, Set the parameters for the firefly algorithm; S1.2 Calculate the brightness corresponding to each firefly; S1.3 Update the firefly's position according to the firefly movement formula model; S1.4 Find and record the brightest firefly; S1.5 When the maximum number of iterations is reached and the firefly's movement distance is less than a preset movement distance threshold, the position of the firefly with the highest brightness is recorded as follows. If the above conditions are met, the position array is obtained; otherwise, repeat S1.2-S1.5; where the position array is the array obtained by the optimal threshold of the two-dimensional image thresholding method in this iteration. The specific steps of S5 include: Normalize the region of interest; Calculate the direction and magnitude of the gradient for all pixels in the region of interest image; Divide the image into A cell units and calculate the gradient histogram for each cell unit; where A is a positive integer greater than 1. Identify key cell units and magnification factors; Recombining B cells from A cells results in a block, and the image formed by all blocks is identical to the original image; where B is a positive integer greater than 1. The features of the cell units within each block are concatenated to obtain the features of that block. The features of all blocks within the image are concatenated to obtain the final second feature vector of the image.

2. The infrared target detection method based on entropy and distance-weighted features in complex backgrounds according to claim 1, characterized in that, The specific steps of S1.1 include: Initialize the firefly population; Set the maximum number of iterations to M; Set the population size of fireflies to N; where M and N are integers greater than or equal to 1; Set the random step size factor to ; Set the light intensity absorption coefficient of the propagation medium to be... ; Set the initial attraction level to ; Initialize the firefly's position.

3. The infrared target detection method based on entropy and distance-weighted features under complex backgrounds as described in claim 1, characterized in that, The specific steps of S2 include: Opening operations are used to remove noise from the foreground region of infrared images; Closing operations are used to repair the foreground region in infrared images.

4. The infrared target detection method based on entropy and distance-weighted features under complex backgrounds as described in claim 1, characterized in that, The specific steps of S3 include: Limiting the shape factor of the connected region reduces its size. The 8-neighborhood method is used to determine whether a set of pixels belongs to a connected component. The set of pixels belonging to the connected components is determined as the final region of interest.

5. The infrared target detection method based on entropy and distance-weighted features in complex backgrounds according to claim 4, characterized in that, The specific steps of S4 include: Classify the 8 pixels in the neighborhood of each region of interest according to distance, and calculate the sum of the local binary pattern values ​​of the center pixel; Calculate the corresponding weights for the calculated local binary pattern values; Calculate the final binary pattern value of the center pixel based on the weights; Calculate the histogram for each region of interest and normalize the histogram. By concatenating the normalized histograms, the first feature vector of the region of interest is obtained.

6. The infrared target detection method based on entropy and distance-weighted features under complex backgrounds according to claim 1, characterized in that, The specific steps of S6 include: The first and second eigenvectors obtained in S4 and S5 are normalized. The normalized first feature vector and the normalized second feature vector are weighted and fused. The final fused feature vector is obtained.

7. The infrared target detection method based on entropy and distance-weighted features in complex backgrounds according to claim 1, characterized in that, The specific steps of S7 include: The fused feature vector obtained in S6 is input into the support vector machine for classification training; The support vector machine is trained C times to obtain the classifier model, where C is a positive integer greater than 1; The extracted feature vectors are classified using a classifier model to obtain the final target detection result.

Images