Infrared small target detection method and device based on image information entropy and multi-scale local contrast measure

Abstract

The present application relates to the field of infrared image processing, and discloses an infrared small target detection method and device based on image information entropy and multi-scale local contrast measure, comprising: (1) constructing a local weighted information entropy operator to obtain a local weighted information entropy image of an original infrared image; (2) using the weighted information entropy image and the structure tensor theory to construct a background prior and a target prior to obtain a local saliency prior; (3) using the original thermal infrared image to design a local contrast coefficient to obtain a multi-scale local contrast image; (4) using the multi-scale local contrast image and the corresponding structure tensor to design an information filter; and (5) fusing the local saliency prior and the information filter to realize the detection of the infrared small target. The present application comprehensively utilizes the weighted image entropy operator, the multi-scale local contrast measure and the structure tensor theory, and can effectively suppress the background, enhance the target and improve the detection performance of the infrared small target.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image processing, and in particular to an infrared small target detection method and device based on image information entropy and multi-scale local contrast. BACKGROUND

[0002] In recent years, infrared search and tracking systems (IRST) have been widely used in many civilian and military fields. As an important part of the infrared search and tracking system, infrared small target detection plays an important role in early warning, precision guidance, and long-range target detection. Due to long-distance imaging, small targets usually lack specific shapes, texture details, and other structural information, presenting a point-like feature and occupying a small proportion of the pixel space in the entire image. At the same time, affected by complex background, clutter, and atmospheric radiation, similar target objects appear in the background, which is easy to cause algorithm false detection. In addition, the image signal-to-clutter ratio is low, and the target is even submerged in the background. Buildings and other interference also have adverse effects on infrared small target detection. Therefore, how to suppress clutter and enhance targets in a complex background to achieve accurate detection of infrared small targets is a challenge currently faced.

[0003] Since the 1980s, domestic and foreign scholars have proposed many infrared small target detection and tracking methods, which can be divided into two categories: Track Before Detect (TBD) and Detect Before Track (DBT). The DBT method takes a single frame image as the processing object, obtains the target detection result of the single frame image, and then predicts and tracks the motion trajectory of the target combined with the time domain information, so the DBT algorithm can also be called a single frame detection algorithm. The detection method based on DBT can be roughly divided into three categories: background suppression-based, human visual system-based (HVS) model, and low-rank sparse representation-based methods. The DBT method can enhance small targets in a single frame image with lower computational complexity and higher detection accuracy, achieving small target detection. However, the DBT method ignores the time information in the image sequence, and it is difficult to distinguish between targets and non-targets in a complex background using only image spatial information. With the development of infrared imaging technology, researchers have begun to focus on the potential information carried in the infrared image sequence, such as temporal features, and more and more DBT methods have been proposed. Unlike the detection strategy of DBT, TBD assumes that the motion trajectory of the target has continuity, predicts the current motion state of all potential targets from the motion characteristics of the target, and uses certain discrimination conditions and the target state in the current frame to judge the reliability of the potential trajectory, thereby completing the detection of small targets in all image frames. Common TBD methods include particle filtering, dynamic programming, etc., but the detection algorithm based on multiple frames has high computational complexity and large hardware requirements.

[0004] With the rapid development of deep learning, more and more researchers have proposed infrared small target detection models based on deep learning. Unlike traditional detection methods, deep learning-based detection methods learn infrared small target representation in a data-driven manner, which can be divided into detection-based methods and pixel-level segmentation-based methods. As a data-driven method, a large amount of training data is essential for feature extraction and learning. However, the labeling cost of a large number of infrared image data is too high. In addition, although the algorithm performance is constantly improving, the phenomenon of deep feature loss still exists, which leads to poor robustness of the algorithm to changing scenes, and cross-layer fusion may even reduce the detection performance. At present, there is still a lack of deep learning-based infrared small target detection algorithm that can achieve good detection effect under complex background and low signal-to-noise ratio conditions. SUMMARY

[0005] In view of the deficiencies in the prior art, the purpose of the present application is to provide an infrared small target detection method and device based on image information entropy and multi-scale local contrast measure, which fully utilizes the characteristics that image information entropy can reflect the gray distribution space features, and the information entropy image can highlight the areas with rich information in the original image and suppress the areas with single information. Through the information entropy image of the infrared image instead of the original infrared image, the local saliency prior in the infrared image can be more effectively extracted. Considering that the local contrast measure can highlight the areas with strong difference, especially the salient areas such as target areas and linear edge areas, a multi-scale local contrast image is designed, and an information filter is constructed on this basis to filter out the invalid information of the local saliency prior, enhance the effective target prior, and realize infrared small target detection.

[0006] To achieve the above purpose, the present application provides the following technical solutions:

[0007] The present application discloses an infrared small target detection method based on image information entropy and multi-scale local contrast measure, characterized by comprising the following steps:

[0008] Step 1): Construct a local weighted information entropy operator E(·) to obtain a local weighted information entropy image E that can reflect the information amount of the thermal infrared image;

[0009] Step 2): Calculate the structure tensor J ι of the local weighted information entropy image E of the original thermal infrared image, and construct the background prior w bp and the target prior w tp , complete the construction of the local saliency prior W l by harmonic averaging of the target prior and the background prior;

[0010] Step 3): using the original thermal infrared image, the local contrast coefficient c(x, y) is calculated by the average value of the pixels in the neighborhood centered at the pixel point (x, y) and the square ring region adjacent to the neighborhood, and a multi-scale local contrast image F(x, y) is obtained;

[0011] Step 4): using the multi-scale local contrast image and the corresponding structure tensor, an edge weighted matrix W edge and a point weighted matrix W point are constructed, and an information filter W filter is designed;

[0012] Step 5): fusing the local saliency prior W l and the information filter W filter , the infrared small target is detected.

[0013] The application further discloses an infrared small target detection device based on image information entropy and multi-scale local contrast quantity for implementing the method, which comprises:

[0014] A local weighted information entropy operator construction module is used to acquire a local weighted information entropy image of an original thermal infrared image, and the local weighted information entropy image is used for image feature region extraction in a local saliency prior generation module;

[0015] A local saliency prior generation module is used to extract an image feature region and construct a target prior and a background prior, and a local saliency prior of an infrared image is obtained by using a harmonic average of the target prior and the background prior;

[0016] A multi-scale local contrast image generation module is used for local contrast coefficient design, and a multi-scale local contrast image obtained is used for information filter construction in an information filter construction module;

[0017] An information filter construction module is used to construct an edge weighted matrix and a point weighted matrix by using the multi-scale local contrast image, and then construct an information filter for filtering invalid information of the local saliency prior;

[0018] A target detection module is used to fuse the local saliency prior and the information filter, and detect an infrared small target;

[0019] A target detection result output module is used to output an infrared small target detection result image.

[0020] The application has the following beneficial effects:

[0021] 1) The present application makes full use of the structural characteristics of small targets and the advantage that image information entropy can reflect the spatial features of gray scale distribution in the image, and an improved weighted image information entropy operator is designed. The weighted information entropy image obtained can highlight the regions with rich information (such as corner and edge regions) in the original image and suppress the regions with single information (such as flat regions). Further, the local saliency prior is extracted by using the structure tensor of the weighted information entropy image to distinguish the flat regions, edge regions and corner regions in the image, and is superior to the local saliency prior extracted by using the original image.

[0022] 2) The present application fully considers that the local contrast measure can highlight the regions with strong differences, especially the salient regions such as target regions and linear edge regions, and a multi-scale local contrast image is designed. In order to further suppress the invalid saliency prior in the local saliency prior due to the strong edge structure, an efficient information filter is constructed by using the multi-scale local contrast image, which suppresses the invalid prior generated by the linear structure such as strong edge in the background, enhances the effective target prior, and improves the background suppression ability and target enhancement ability of the detector. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The overall flowchart of the embodiment of the small target detection method for thermal infrared images of the present application is shown in the figure.

[0024] Figure 2 The structural schematic diagram of the infrared small target detection device of the present application is shown in the figure.

[0025] Figure 3 The example frame image of the thermal infrared image sequence for experimental test is shown in the figure.

[0026] Figure 4 The local weighted entropy image of the example frame is shown in the figure.

[0027] Figure 5a 、 Figure 5b and Figure 5c are the comparison diagrams of the small target detection results after detection by different methods for infrared images and the corresponding three-dimensional gray scale visualization images. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application will be described in detail below with specific examples. The specific examples are described below to simplify the present application. However, it should be recognized that the present application is not limited to the described examples, and various modifications of the present application are possible without departing from the basic principles, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

[0029] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of the present application.

[0030] As shown in the basic step flowchart of the infrared small target detection method based on image information entropy and multi-scale local contrast measure of the present application, the method mainly includes the following steps: Figure 1

[0031] Step 1: Construct a local weighted information entropy operator E(·) to obtain a local weighted information entropy image E reflecting the information quantity of a thermal infrared image;

[0032] In the embodiment, for a 7×7 neighborhood M centered at (x, y), it contains m different pixel values f i (i=1, 2, …, m), and the corresponding probability density function is P i (i=1, 2, …, m). The gray value difference |f i -f(x, y)| 4 (i=1, 2, …, m) in the neighborhood M is taken as a weight to weight the information entropy of the neighborhood M. The weighted information entropy operator e(·) is:

[0033]

[0034] The local standard deviation σ std (x, y) is introduced to measure the uniformity of the image texture in the neighborhood M, enhance the target saliency and suppress the edge structure. Finally, the local weighted information entropy operator E(·) is designed as follows:

[0035]

[0036] Wherein, f(i, j) represents the gray value at position (i, j), and ε1 is 10 -2 to prevent the denominator from being zero.

[0037] Thus, the local weighted information entropy image E reflecting the information quantity of the thermal infrared image is obtained.

[0038] Step 2: Use the local weighted information entropy image E of the original thermal infrared image to calculate the structure tensor J ι , construct the background prior w bp and the target prior w tp , complete the construction of the local saliency prior W l by the harmonic average of the target prior and the background prior.

[0039] Calculate the structure tensor J​ι As follows:

[0040]

[0041] where G ι is a Gaussian kernel function with variance σ, in this example, a Gaussian kernel function with standard deviation 3 and size 3x3, E σ is a Gaussian smoothing filter with variance σ, in this example, E σ is a Gaussian smoothing filter with standard deviation 3 and size 3x3, "*" is a two-dimensional convolution operator, is a Kronecker product, ▽ is a gradient operator, and

[0042] The structure tensor J ι of the locally weighted information entropy image E contains two eigenvalues λ1and λ2, which can be used to characterize the image structure features in E: i) edges, λ1>> λ2≈0; ii) corners, λ1≥ λ2>>0; iii) planes, λ1≈ λ2≈0; and further construct the background prior w bp = max(λ1, λ2) and the target prior ε2=10 -4 The local saliency prior W l ' is obtained by harmonic averaging of the target prior and the background prior, and is normalized:

[0043]

[0044]

[0045] where ε2and ε3are normal numbers to prevent the denominator from being zero, w min and w max are the minimum and maximum values of W l ', respectively.

[0046] Step 3: Using the original thermal infrared image, the local contrast coefficient c(x, y) is calculated by the average value of the pixels in the neighborhood centered at pixel point (x, y) and the square ring region adjacent to the neighborhood, and a multi-scale local contrast image F(x, y) is obtained.

[0047] The neighborhood Ω g is defined as a neighborhood centered at pixel point (x, y) with size (2g+1)x(2g+1), and the calculation formula is:

[0048] Ω g = {(u, v) | max(|u-x|, |v-y|)≤g} (6)

[0049] where g represents the neighborhood length coefficient;

[0050] obtaining a neighborhood Ω of a pixel point (x, y) g and the average gray value c of n adjacent square ring regions adjacent to the neighborhood i , i = 0, 1, …, n:

[0051]

[0052]

[0053] wherein f(u, v) represents a gray value at a pixel point (u, v), N g represents a total number of pixels in the neighborhood Ω g , (u, v) ∈ Ω g represents that the pixel point (u, v) belongs to the neighborhood Ω g , (u, v) ∈ R i represents that the pixel point (u, v) belongs to the i-th square ring, i.e., (u, v) ∈ Ω g+i and

[0054] Therefore, a multi-scale local contrast image F(x, y) is constructed as follows:

[0055] F(x, y) = f(x, y) c(x,y) (9)

[0056]

[0057] wherein c(x, y) is a local contrast coefficient of the pixel point (x, y).

[0058] Step 4: using the multi-scale local contrast image and the corresponding structure tensor, an edge weighted matrix W edge and a point weighted matrix W point are constructed, and an information filter W filter is designed;

[0059] Specifically, the eigenvalues of the structure tensor J ι ′ of each pixel point (x, y) in the multi-scale local contrast image F(x, y) are λ1′, λ2′, the determinant is Φ d = λ1′λ2′, and the trace is Φ t = λ1′+ λ2′; considering that the determinant and the trace of the structure tensor of a corner / planar region in the image are large / small, and the determinant of a linear edge region is small and the trace is large, it can be represented as:

[0060]

[0061] wherein Φ t (x, y) represents the trace of the structure tensor at the pixel point (x, y), Φd (x,y) represents the determinant of the structure tensor at pixel point (x,y), d2>=d1, and t1, t2, d1, d2 are threshold values, and satisfy:

[0062]

[0063] wherein ψ1, ψ2, ψ3, ψ4 are empirical values 0.11, 0.12, 0.01, 0.01, and is the trace and determinant value after removing the infinite large;

[0064] In order to further enhance the point feature target and suppress the strong edge of the line feature, an edge weighting matrix W edge and a point weighting matrix W point are constructed.

[0065]

[0066]

[0067] The edge weighting matrix W edge and the point weighting matrix W point designed prior information filter W filter are:

[0068] W filter = W edge (x,y) + γW point (x,y) (15)

[0069] wherein γ is a weight coefficient.

[0070] Step 5: fuse the local saliency prior W l and the information filter W filter to detect the infrared small target.

[0071] In the step 5), the infrared small target detection result graph T is obtained by fusing the local saliency prior W l and the information filter W filter .

[0072] T = W l · W filter = W l · (W edge + γW point ) (16).

[0073] Corresponding to the aforementioned one embodiment of the infrared small target detection method based on image information entropy and multi-scale local contrast measure, the present application further provides one embodiment of an infrared small target detection device based on image information entropy and multi-scale local contrast measure.

[0074] Figure 2 For a block diagram of an infrared small target detection device based on image information entropy and multi-scale local contrast measure according to an exemplary embodiment, as shown in Figure 2 The device comprises:

[0075] A local weighted information entropy operator construction module is configured to obtain a local weighted information entropy image of the original infrared image, and the local weighted information entropy image is used for image feature region extraction in a local saliency prior generation module.

[0076] The local saliency prior generation module is configured to extract image feature regions and construct target prior and background prior, and obtain local saliency prior of the infrared image by harmonic average of the target prior and the background prior.

[0077] A multi-scale local contrast image generation module is configured to design local contrast and obtain a multi-scale local contrast image, which is used for an information filter construction module to construct an information filter.

[0078] The information filter construction module is configured to construct an edge weighted matrix and a point weighted matrix by using the multi-scale local contrast image, and further construct an information filter for filtering invalid information of the local saliency prior.

[0079] A target detection module is configured to fuse the local saliency prior and the information filter to realize small target detection.

[0080] A target detection result output module is configured to output an infrared small target detection result image.

[0081] In a specific implementation of the present application, an application expansion module is further included, which is configured to use image information entropy and multi-scale local contrast measure to perform infrared small target detection and tracking.

[0082] As to the device in the above-mentioned embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and thus will not be described in detail here.

[0083] For the device embodiment, since it basically corresponds to the method embodiment, the relevant part can be seen from the part of the method embodiment. The device embodiment described above is only illustrative, and each module in the device is a logical functional division. In actual implementation, there can be another division manner, for example, multiple modules can be combined or integrated into another unit. In addition, the connection between the displayed or discussed modules can be a communication connection through some interface, which can be electrical or other forms. According to actual needs, part or all of the modules can be selected to achieve the purpose of the scheme of the present application. Those skilled in the art can understand and implement without paying creative labor. The following takes the disclosed real thermal infrared image sequence as an example to illustrate the specific implementation manner, so as to embody the technical effect of the present application. The specific steps in the embodiment will not be repeated.

[0084] Embodiment

[0085] The drawings shown in the embodiment of the present application can make the purpose, technical scheme and advantage of the present application more clear and understandable. It should be noted that the specific embodiments described herein are only used to explain the present application and not to limit the present application. Any equivalent replacement, improvement, etc. made within the method idea and principle provided by the present application should be included in the protection scope of the present application.

[0086] Next, taking the disclosed infrared image sequence as the research object, the infrared small target detection algorithm verification is carried out. In order to evaluate the small target detection result from the qualitative and quantitative angles, the intuitive result graph of small target detection and the 3D-ROC evaluation index system including AUC TD , AUC BS , AUC SNPR , AUC TDBS and AUC ODP , etc. are used to comprehensively evaluate the small target detection result from the target detection ability, background suppression ability and detector effectiveness, and the specific definitions are as follows:

[0087] AUC (D,F) represents the AUC value of the (P D , P F ) curve, representing the effectiveness of the detector,

[0088] AUC (D,F) ∈[0,1]

[0089] AUC (D,τ) represents the AUC value of the (P D , τ) curve, representing the target detection ability of the detector,

[0090] AUC (D,τ) ∈[0,1]

[0091] AUC (F,τ) AUC D ,τ) curve, characterizing the background suppression ability of the detector,

[0092] AUC (F,τ) ∈[0,1]

[0093] AUC TD characterizing the joint detection ability of the detector,

[0094] AUC TD = AUC (D,F) + AUC (D,τ) )∈[0,2]

[0095] AUC BS characterizing the joint background suppression ability of the detector,

[0096] AUC BS = AUC (D,F) - AUC (F,τ) )∈[-1,1]

[0097] AUC TDBS characterizing the combined target detection and background suppression ability of the detector,

[0098] AUC TDBS = AUC (D,τ) - AUC (F,τ) )∈[-1,1]

[0099] AUC SNPR characterizing the signal-to-noise ratio of the detector,

[0100]

[0101] AUC ODP characterizing the total detection probability of the detector,

[0102] AUC ODP = AUC (D,F) + AUC (D,τ) - AUC (F,τ) ∈[-1,2]

[0103] In general, the 3D-ROC evaluation index system is divided as follows:

[0104] (i) Target detection ability (TD): AUC (D,τ) , AUC TD ;

[0105] (ii) Background suppression ability (BS): AUC (F,τ) , AUC BS , AUC SNPR ;

[0106] (iii) Detector effectiveness: AUC (D,F) ,AUC TDBS ,AUC ODP .

[0107] For a public real 100-frame image thermal infrared image sequence, the characteristics are as follows: the aircraft in the background moves slowly, the local contrast of the target-background is low, and the background contains suspected target objects and many linear edge structures, and the image size is 256*256. An example frame of the infrared image sequence is shown in FIG. 1, and the corresponding local weighted entropy image is shown in FIG. 2, and the salient regions such as targets and edges can be highlighted. Table 1 is a quantitative index of the small target detection results of the thermal infrared image sequence by using TLLCM, IPI, NIRAM, NIRPS, RIPT, LogTFNN, WSNM-STIPT, ASTTV-NTLA and the proposed method, and the bold and underlined values represent the optimal AUC results and the suboptimal AUC results, respectively. Figure 3 Figure 4 Table 1 is a quantitative index of the small target detection results of the thermal infrared image sequence by using TLLCM, IPI, NIRAM, NIRPS, RIPT, LogTFNN, WSNM-STIPT, ASTTV-NTLA and the proposed method, and the bold and underlined values represent the optimal AUC results and the suboptimal AUC results, respectively.

[0108] Table 1 is a quantitative index of the small target detection results of the thermal infrared image sequence by using TLLCM, IPI, NIRAM, NIRPS, RIPT, LogTFNN, WSNM-STIPT, ASTTV-NTLA and the proposed method, and the bold and underlined values represent the optimal AUC results and the suboptimal AUC results, respectively.

[0109] Method AUC (D,F) ]] AUC (D,τ) ]] AUC (F,τ) ]] AUC TD ]] AUC BS ]] AUC SNPR ]] AUC TDBS ]] AUC ODP ]] Invention 1.0000 1.0000 0.0051 2.0000 0.9949 1.9636e2 0.9949 1.9949 TLLCM 0.9981 0.3513 0.0055 1.3494 0.9926 6.4188e1 0.3458 1.3439 IPI 1.0000 1.0000 0.0762 2.0000 0.9238 1.3118e1 0.9238 1.9238 NRAM 1.0000 0.9988 0.0055 1.9988 ]]> ​ 0.9945 1.8171e2 0.9933 1.9933 NIRPS 0.9984 ]]> ​ 0.5097 0.0071 1.5081 0.9913 7.2256e1 0.5026 1.5010 RIPT 1.0000 1.0000 0.0057 2.0000 0.9943 1.7439e2 0.9943 ]]> ​ 1.9943 ]]> ​ LogTFNN 1.0000 0.9993 ]]> ​ 0.0137 1.9993 0.9863 7.2988e1 0.9856 1.9856 WSNM-STIPT 1.0000 0.9610 0.0053 ]]> ​ 1.9610 0.9947 ]]> ​ 1.8295e2 ]]> ​ 0.9557 1.9557 ASTTV-NTLA 1.0000 0.9970 0.0683 1.9970 0.9317 1.4602e1 0.9287 1.9287

[0110] Figure 5a 、 Figure 5b 、 Figure 5c For the small target detection results of the thermal infrared image sequence by using TLLCM, IPI, NIRAM, NIRPS, RIPT, LogTFNN, WSNM-STIPT, ASTTV-NTLA and the proposed method, and the corresponding three-dimensional gray visualization images, it can be seen that NIRPS and LogTFNN and other comparison algorithms are easily affected by edges, have a high false alarm, and the proposed method not only has good target detection performance, but also has excellent background suppression capability. From the quantitative detection results of the eight evaluation indexes shown in Table 1, it can be seen that the method proposed in the present application is better than the comparison methods in the eight 3D-ROC derived evaluation indexes, reflecting that the target detection capability, background suppression capability and detector effectiveness of the proposed algorithm are better than those of the existing advanced small target detection algorithm. Comprehensive qualitative analysis and quantitative analysis show that the thermal infrared small target detection method proposed in the present application has excellent thermal infrared small target detection capability, and the target enhancement capability and the background suppression capability are significantly superior.

[0111] ​The drawings of the embodiments of the present application can make the purpose, technical scheme and advantages of the present application more clear. It should be noted that the specific embodiments described herein are only used to explain the present application and not to limit the present application. Any equivalent replacement, improvement, etc. within the method idea and principle provided by the present application should be included in the protection scope of the present application.

Claims

1. An infrared small target detection method based on image information entropy and multi-scale local contrast measure, characterized in that, The method comprises the following steps: Step 1): constructing a local weighted information entropy operator E(·) to obtain a local weighted information entropy image E reflecting the information quantity of the thermal infrared image; The step 1) is specifically: For a neighborhood M of N x N size centered at (x, y), it contains m different gray values f i , i = 1, 2, …, m, the corresponding probability density function is P i , i = 1, 2, …, m, the difference of gray values in neighborhood M is |f i -f(x, y)| n As the weight, n represents the defined exponential coefficient, the information entropy of neighborhood M centered at (x, y) is weighted, and the weighted information entropy operator e(·) is: By introducing a local standard deviation σ std (x, y) to measure the uniformity of image texture in the neighborhood M, enhance the target saliency and suppress the edge structure, the final designed local weighted information entropy operator E(·) is as follows: Wherein, f(i,j) represents the gray value at the position (i,j), and ε1 is a normal number to prevent the denominator from being zero; Thus, the local weighted information entropy image E reflecting the information quantity of the thermal infrared image is obtained; Step 2): using the local weighted information entropy image E of the original thermal infrared image, calculating the structure tensor J ι , and constructing the background prior w bp , and the target prior w tp , and completing the construction of the local saliency prior W l by harmonic mean of the target prior and the background prior; Step 3): using the original thermal infrared image, the local contrast coefficient c(x,y) is calculated by the pixel average value of the neighborhood with the pixel point (x,y) as the center and the square ring region adjacent to the neighborhood, and a multi-scale local contrast image F(x,y) is obtained; Step 4): Constructing edge-weighted matrix W using multiscale local contrast images and corresponding structure tensors edge and point-weighted matrix W point , designing information filter W filter ; Step 5): fusing local saliency priors W l and information filter W filter for detecting infrared small targets.

2. The method of claim 1, wherein the infrared small target detection method based on image information entropy and multi-scale local contrast measure is characterized in that, The step 2) is specifically: Computing the structure tensor J of the locally weighted information entropy image E ι As follows: where G ι is a Gaussian kernel function with variance σ, E σ is a Gaussian smoothing filter with variance σ, and "*" is a two-dimensional convolution operator, is the Kronecker product, and ▽ is the gradient operator, and Structure tensor J of the locally weighted information entropy image E ι With two eigenvalues λ1and λ2, which can be used to depict the image structure characteristics in E: i) edge, λ1>> λ2≈ 0; ii) corner, λ1≥ λ2>> 0; iii) plane, λ1≈ λ2≈ 0; and further construct the background prior w bp = max(λ1, λ2) and the target prior The local saliency prior W is obtained by the harmonic average of the target prior and the background prior l ′, and normalize it: where ε2and ε3are normal numbers, preventing the denominator from being zero, w min and w max are the minimum and maximum values of W l ′, respectively.

3. The method of claim 1, wherein the method further comprises: The step 3) is specifically: Definition of neighborhood Ω g For the neighborhood centered at pixel point (x, y) with size (2g+1) x (2g+1), the calculation formula is: Ω g = {(u, v) | max(|u - x|, |v - y|) < g} (6) Wherein, g represents the neighborhood side length coefficient; Obtain the gray mean value c of the neighborhood Ω of the pixel point (x, y) and the n adjacent square ring regions adjacent to the neighborhood g i i = 0, 1, …, n:​ where f(u, v) denotes the gray value at pixel point (u, v), N g denotes the total number of pixels in the neighborhood Ω g denotes the total number of pixels in the neighborhood Ω g+i denotes the total number of pixels in the neighborhood Ω g+i denotes the total number of pixels in the neighborhood Ω g+i-1 denotes the total number of pixels in the neighborhood Ω g+i-1 denotes the total number of pixels in the neighborhood Ω g denotes that pixel point (u, v) belongs to the neighborhood Ω g denotes that pixel point (u, v) belongs to the i-th square ring, i.e. i denotes that pixel point (u, v) belongs to the i-th square ring, i.e. g+i denotes that pixel point (u, v) belongs to the i-th square ring, i.e. Thus, the multi-scale local contrast image F(x,y) is constructed as: F(x, y) = f(x, y) c(x,y) (9) Wherein, c(x,y) is the local contrast coefficient of the pixel point (x,y).

4. The method of claim 1, wherein the method further comprises: In step 5), the local saliency prior W is fused l and the information filter W filter to obtain the infrared small target detection result image T: T = W l • W filter = W l • (W edge + γW point ) (16) γ is a weight coefficient.

5. An infrared small target detection device based on image information entropy and multi-scale local contrast measure for implementing the method of claim 1, characterized in that, It comprises: A local weighted information entropy operator construction module, which is used to obtain a local weighted information entropy image of the original thermal infrared image, and the local weighted information entropy image is used for image feature region extraction in a local saliency prior generation module; A local saliency prior generation module, which is used for image feature region extraction and construction of target prior and background prior, and a local saliency prior of the infrared image is obtained by using the harmonic average of the target prior and the background prior; A multi-scale local contrast image generation module, which is used for local contrast coefficient design, and a multi-scale local contrast image obtained is used for construction of an information filter in an information filter construction module; An information filter construction module, which is used to construct an edge weighted matrix and a point weighted matrix by using the multi-scale local contrast image, and then construct an information filter for filtering invalid information of the local saliency prior; A target detection module, which is used to fuse the local saliency prior and the information filter to perform infrared small target detection; A target detection result output module, which is used to output an infrared small target detection result image.

Images