Weak and Small Target Detection Method and System Based on Polarization Consistency Local Contrast

Abstract

The present invention discloses a method and system for detecting small and weak targets based on polarization consistency local contrast. The method includes: obtaining a grayscale image of a small and weak target and corresponding multiple polarization images; implementing fusion denoising of the grayscale image and the polarization images based on a guided filter algorithm to obtain an output image; traversing the entire output image by using the mechanism of local contrast and combining a joint polarization enhancement coefficient to obtain a polarization degree saliency map of the whole image; in the polarization degree saliency map, using an adaptive threshold to segment the dim and weak targets from the background to complete the detection of small and weak targets. By using the present invention, the problem of detecting infrared small and weak targets in a complex background can be solved, and the accurate detection of infrared small and weak targets can be achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of optoelectronic detection and imaging, and in particular relates to a method and system for detecting small and weak targets based on polarization consistency local contrast. Background Technique

[0002] The enhanced detection of small and weak targets is one of the key technologies in application fields such as guidance, early warning, airborne or spaceborne monitoring, and anti-drone, and is also an important research direction for traditional small and weak target detection.

[0003] However, traditional small and weak target detection can only obtain targets under specific backgrounds, that is, by having an obvious gray-scale difference between the background gray scale and the target gray scale, extracting the difference information between each pixel point in the image and its neighboring pixel points, and the target can be successfully screened out.

[0004] For example, the Chinese patent document with the publication number CN112396061A discloses an Otsu target detection method based on target gray-scale tendency weighting, including the following steps: Step S1: Perform gray-scale processing on the image to be measured to obtain the gray-scale histogram of the image to be measured; Step S2: Determine the expected threshold according to the obtained gray-scale histogram of the image to be measured; Step S3: Calculate the occurrence probabilities and gray-scale means of foreground and background pixel points according to the expected threshold; Step S4: Select the gray-scale tendency coefficient to be changed according to the expected threshold. If the expected threshold is closer to the high-gray-scale area, select the high-tendency coefficient α, and vice versa, if it is closer to the low-gray-scale area, select the low-tendency coefficient β; Step S5: Use the improved Otsu between-class variance formula for gradient valley value weighting to obtain the threshold that maximizes the variance and realize the separation of the target and the background.

[0005] The Chinese patent document with the publication number CN117422740A discloses an infrared moving target detection and processing method, including: taking an infrared moving target picture, first performing filtering processing on the picture; then using a method based on the change of gray-scale value to segment the target area and remove the shadow; performing binary processing and dilation processing on the segmented target area; then performing contour finding processing; after the target contour screening operation, identify and lock the position of the infrared target in the picture.

[0006] However, in the face of various complex scenarios such as rain, snow, haze, sandstorm and other complex weather, scattering media such as smoke and underwater, or non-cooperative targets such as camouflaged targets and transparent objects, these current traditional small and weak target detection methods will lose their detection capabilities.

[0007] To solve the above problems, currently mainly the local contrast method based on the characteristics of the human visual system is proposed, that is, according to the characteristics of human vision, quickly obtain the significant area where the target exists from the complex background and eliminate the invalid background information.

[0008] However, the local contrast method faces the following two problems: First, restricted by the definition of the local contrast principle, only the difference information between the target and the neighborhood background is extracted during the calculation process, and no enhancement operation is performed on the real target. When the target is very small, it is easy to cause missed detection. Second, no background reduction operation is performed during the calculation. When there is a large area of high-brightness background area in the image, false detection is likely to occur during the image enhancement operation. Summary of the Invention

[0009] The present invention provides a method and system for detecting small and weak targets based on polarization consistency local contrast, which can solve the problem of detecting small and weak infrared targets in complex backgrounds and achieve accurate detection of small and weak infrared targets.

[0010] The method for detecting small and weak targets based on polarization consistency local contrast includes the following steps:

[0011] (1) Obtain the grayscale image of the small and weak target and the corresponding multiple polarization images;

[0012] (2) Based on the guided filter algorithm, perform fusion denoising on the grayscale image and the polarization images to obtain an output image;

[0013] (3) Using the mechanism of local contrast and combining the joint polarization enhancement coefficient, traverse the entire output image to obtain the polarization degree saliency map of the whole image;

[0014] (4) In the polarization degree saliency map, use an adaptive threshold to segment the dim target and the background to complete the detection of small and weak targets.

[0015] In step (1), by rotating the polarizer to change the angle of the analyzer, the rotation angles are 0°, 45°, 90°, and 135° respectively, and the corresponding four polarization images are , , , .

[0016] The specific process of step (2) is as follows:

[0017] Take the grayscale image as the guiding image and the polarization image as the image to be filtered for guided filtering; for non-target areas, the gradient of the grayscale image is small, and the output polarization image is a smoothed result, and the noise is suppressed; for target areas, the gradient of the grayscale image is large, and the target in the output polarization image is retained.

[0018] The specific process of step (3) is as follows:

[0019] Slide the sliding window from left to right and from top to bottom to traverse the entire image area, which is divided into several sliding window areas;

[0020] Each sliding window region is regarded as a sub-region, and each sub-region is divided into a central pixel region and a surrounding pixel region; obtain the maximum polarization degree in the central pixel region , the average polarization degree and the average polarization degree of the surrounding pixel region , where

[0021] represents the number of the surrounding pixel region; ;

[0022] Introduce the polarization degree saliency map of the joint polarization enhancement coefficient as:

[0023] ;

[0024] In the formula, represents the serial number of the sliding window region, represents the polarization saliency map of the th sliding window region introduced with the joint polarization enhancement coefficient, represents the gray difference ratio between the central pixel region and the th surrounding pixel region.

[0025] Traverse the entire image to obtain the polarization degree saliency map of the entire image.

[0026] Construct the joint polarization enhancement coefficient according to the polarization degree and polarization consistency, and the formula is as follows:

[0027] ;

[0028] In the formula, is the polarization degree in the entire sliding window region; is the polarization standard deviation in the entire sliding window region to represent the polarization consistency.

[0029] and The calculation formulas of

[0030] are as follows:

[0031] ;

[0032] In the formula, represents the light intensity information, and represent the linear polarization information, , and are obtained through the following formulas:

[0033] ;

[0034] Wherein, represents the vector obtained from multiple polarization images, represents the light intensity information, and represent the linear polarization information, represents the circular polarization information. Since the ground-based polarization detection system does not collect circular polarization information, thus the component is 0; , , , are the four polarization images corresponding to the analyzer angles of 0°, 45°, 90°, and 135° of the rotating polarizer, respectively.

[0035] In step (5), the adaptive threshold is defined as:

[0036] ;

[0037] Wherein, and are both the mean and variance of the polarization saliency map, is an adjustable constant parameter. When the value is greater than the threshold , this pixel is the target, otherwise it is the background.

[0038] The present invention also provides a small and weak target detection system based on polarization consistency local contrast, which is characterized by including a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, it is used to implement the above-mentioned small and weak target detection method.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention obtains the polarization image and grayscale image of the small and weak target through the polarization detection system, realizes polarization image noise reduction by fusing the polarization image and grayscale image based on the guided filter algorithm, and obtains the joint polarization coefficient through the characteristics of polarization degree and polarization direction consistency; enhances the polarization image target through the target contrast of the joint polarization coefficient; and realizes the detection of small and weak targets through the adaptive threshold, and finally can realize the enhanced detection of infrared small and weak targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is the flow chart of the small and weak target detection method based on polarization consistency local contrast of the present invention.

[0042] Figure 2 Schematic diagram of calculating the polarization saliency map in the embodiment of the present invention. Detailed implementation manners

[0043] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be noted that the following embodiments are intended to facilitate the understanding of the present invention and do not limit it in any way.

[0044] Considering that the polarization information is independent of the radiation intensity information, it can highlight high-frequency information such as target edges and details in complex backgrounds, and the polarization information is determined by factors such as the object material and the light irradiation angle. Therefore, it is more suitable for detecting small and weak targets in complex backgrounds. Based on these characteristics of the polarization information, the polarization information is applied to the local contrast method, and the enhanced detection of small and weak targets is realized through the polarization difference between the target and the background.

[0045] As Figure 1 shown, the method for detecting small and weak targets based on polarization consistency local contrast includes the following steps:

[0046] Step 1: Obtain the grayscale image and polarization image of the dim target.

[0047] First, obtain the grayscale image of the small and weak target through the polarization detection system, and then rotate the polarizer. By changing the angle of the analyzer, the rotation angles are 0°, 45°, 90°, and 135° respectively, and the polarization images of the same scene collected correspondingly are , , , .

[0048] Step 2: Realize the fusion and noise reduction of the polarization image and the grayscale image based on the guided filter algorithm.

[0049] Filtering the image with a filter having a smoothing effect can suppress the noise in the image, but this will also reduce the contrast of the small and weak target. Therefore, the ideal processing effect is to smooth the noise in the image without changing the contrast between the small and weak target and the surrounding background.

[0050] Combining the characteristics of the polarization image and the grayscale image, use the guided filter algorithm to fuse the polarization image and the grayscale image to realize the noise reduction of the polarization image. The guided filter is an edge-preserving filter, which requires an input guide image and an image to be filtered, and then assumes that there is a local linear relationship between the output image and the guide image, that is, the following formula holds:

[0051] ;

[0052] where and represent the output image and the guide image a certain pixel in , represents the area of the sliding window, represents the window and the linear relationship parameters satisfied by the guidance image and the output image in it.

[0053] The guided filtering algorithm defines the input image and the output image error function:

[0054] ;

[0055] ;

[0056] Among them, represents the input and output error function, is a certain pixel in the input image, and the coefficient corresponding to the minimum error function is solved :

[0057] ;

[0058] ;

[0059] Among them, and respectively represent the mean and variance of the guidance image in the window , represents the calculation of and covariance, represents the gray mean of the input image in the window . Finally, the output of any pixel in the window is calculated according to the above formula, and then the calculation results of all windows containing the pixel point are averaged to obtain the output image

[0060] ;

[0061] Among them, , . It can be deduced from the derivation that the guided filtering can input different regions of the guidance image according to the gradient of the guidance image. If the input is a smooth region of the guidance image, the calculated is very small, is close to , so the output image is the result of smoothing the input image; if the input is a region with a large gradient of the guidance image, the guidance image has a large variance, and the calculated is close to 1,​Close to 0, so the output image also has a large gradient.

[0062] Using the grayscale image as the guiding image and the polarization image as the image to be filtered for guided filtering. For non-target regions, the gradient of the grayscale image is small, and the output polarization image is a smoothed result with noise suppressed; for target regions, the gradient of the grayscale image is large, and the target of the output polarization image is retained. Therefore, using guided filtering to process the polarization image can effectively suppress noise while retaining the target of the polarization image.

[0063] Step 3, obtain the joint polarization enhancement coefficient.

[0064] The polarization detection system obtains the grayscale image of the scene by changing the angle of the analyzer, and the rotation angles are 0°, 45°, 90°, and 135° respectively. The corresponding acquired images are 、 、 、 , and through these four images, the vector can be calculated:

[0065] ;

[0066] In the formula, represents the vector, represents the light intensity information, and represent the linear polarization information, represents the circular polarization information. Since the ground-based polarization detection system does not collect circular polarization information, the component is 0. The

[0067] vector can represent the entire polarization state of the light ray, and the polarization state information is often characterized by the degree of polarization and the polarization angle. The degree of polarization represents the ratio of the intensity of the completely polarized light in the light beam to the total light intensity, and the polarization angle represents the direction of polarization of the light beam. The calculation methods are as follows:

[0068] ;

[0069] Polarization consistency mainly describes the consistency of the polarization direction and refers to the degree of change of the polarization angle within the detection area. Specifically, the polarization angle changes little within the target area, and the polarization directions are relatively unified, that is, the target area has higher polarization consistency. The polarization distribution in the background area is relatively random, resulting in a large change in the polarization angle within the area, that is, the background area has lower polarization consistency. Among them, the degree of polarization and the polarization consistency are used to construct the joint polarization enhancement coefficient and the calculation is as follows:

[0070] ;

[0071] Step 4, the target contrast enhancement method combining polarization degree and polarization angle.

[0072] According to the analysis of the polarization characteristics of the target and the background, it can be seen that there are obvious differences in both the polarization degree and the polarization direction consistency between the target and the background. Therefore, a method using two parameters, polarization degree and polarization consistency, is proposed to detect dim targets. Local contrast is the contrast between the target and the background, which can better highlight the target and suppress the background, making the target more prominent during the detection process. Therefore, based on the mechanism of local contrast, a polarization degree saliency map enhanced by combining two parameters of polarization degree and polarization direction consistency is proposed.

[0073] The polarization degree saliency map is obtained by calculating the local contrast, and the local contrast evaluates the contrast difference between the central pixel and the surrounding pixels through a sliding window. As shown in shown, represents the central pixel area, 1-8 represent the surrounding pixel areas. The sliding window is slid from left to right and from top to bottom to traverse the entire image area, which is divided into several sliding window areas.

[0074] In each sliding window area, the sub-area is divided into the central pixel area , and the surrounding pixel area (1-8).

[0075] The maximum polarization degree of the central pixel area is:

[0076] ;

[0077] Among them, is the maximum polarization degree in area T, is the polarization degree value of each pixel in area T.

[0078] The average polarization degree of the sub-area is:

[0079] ;

[0080] Among them, is the average polarization degree in the sub-area, is the polarization degree value of each pixel in the sub-area. is the average polarization degree in area T, is the average polarization degree in the surrounding pixel area.

[0081] Define the difference in polarization degree and polarization direction consistency between the central pixel area and the background area as the influencing factor of local contrast, and the combined polarization enhancement coefficient expression is:

[0082] ;

[0083] Among them, is the degree of polarization in the entire window area, is the standard deviation of polarization in the entire window area.

[0084] Therefore, the saliency map of the degree of polarization introducing the joint polarization enhancement coefficient is:

[0085] ;

[0086] Traverse the entire image to obtain the saliency map of the degree of polarization of the entire image.

[0087] Step 5, detecting small and weak targets based on an adaptive threshold.

[0088] The saliency map of the degree of polarization can highlight small and weak targets more prominently. Therefore, an adaptive threshold is used to segment small and weak targets from the background, and the threshold is defined as:

[0089] ;

[0090] Among them, and are both the mean and variance of the polarization saliency map, is an adjustable constant parameter. When the value is greater than the threshold , this pixel is a target, otherwise it is the background.

[0091] Based on the above method, the enhancement detection of small and weak targets based on polarization consistency local contrast can be realized. By continuously optimizing and adjusting parameters, the detection of small and weak targets in different scenarios can be achieved.

[0092] Based on the same inventive principle, this embodiment also provides a small and weak target detection system based on polarization consistency local contrast, including a memory and one or more processors. When the one or more processors execute the executable code stored in the memory, it is used to implement the small and weak target detection method mentioned in the above embodiment.

[0093] The above embodiments have described the technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modification, supplement, and equivalent replacement made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A small target detection method based on polarization consistency local contrast, characterized in that: The following steps are involved: (1) Obtain the grayscale image of the dim target and the corresponding multiple polarization images; (2) Based on the guided filtering algorithm, the grayscale image and polarization image are fused and denoised to obtain the output image; (3) Using the mechanism of local contrast and the joint polarization enhancement coefficient, the entire output image is traversed to obtain the polarization saliency map of the entire image. The specific process is as follows: Slide the sliding window from left to right and from top to bottom to traverse the entire image area and divide it into several sliding window areas; Each sliding window area is regarded as a sub-area, and each sub-area is divided into a central pixel area. and surrounding pixel areas; get the central pixel area Medium polarization maximum , mean polarization degree And the average polarization value of the surrounding pixel area , Indicates the number of the surrounding pixel area; Constructing joint polarization enhancement coefficient based on degree of polarization and polarization consistency , the formula is as follows: ； In the formula, is the polarization degree in the entire sliding window area; is the polarization standard deviation in the entire sliding window area, which represents the polarization consistency; Introducing the joint polarization enhancement factor The polarization degree significance map is: ； In the formula, Indicates the sequence number of the sliding window area, Indicates The polarization saliency map of the joint polarization enhancement coefficient is introduced in each sliding window region. Indicates the center pixel area and the Grayscale difference ratio of surrounding pixel areas; Traverse the entire image to obtain the polarization saliency map of the entire image; (4) In the polarization saliency map, an adaptive threshold is used to segment the dim target from the background to complete the dim target detection.

2. The method for detecting small targets based on polarization consistency local contrast according to claim 1, characterized in that: In step (1), the polarizer is rotated to change the angle of the analyzer. The rotation angles are 0°, 45°, 90°, and 135°, respectively. The corresponding four polarization images are: , , , .

3. The method for detecting small targets based on polarization consistency local contrast according to claim 1, characterized in that: The specific process of step (2) is as follows: The grayscale image is used as the guide image and the polarization image is used as the image to be filtered for guided filtering; for the non-target area, the gradient of the grayscale image is small, the output polarization image is the result of smoothing, and the noise is suppressed; for the target area, the gradient of the grayscale image is large, and the target is retained in the output polarization image.

4. The method for detecting small targets based on polarization consistency local contrast according to claim 1, characterized in that: and The calculation formula is as follows: ； ； In the formula, Indicates light intensity information, and Represents linear polarization information, , and Obtained by the following formula: ； In the formula, Represents the value obtained by multiple polarization images Vector, Indicates light intensity information, and Represents linear polarization information, Represents circular polarization information. Since the ground-based polarization detection system does not collect circular polarization information, The weight is 0; , , , These are the four polarization images corresponding to the analysis angles of the rotating polarizer being 0°, 45°, 90°, and 135°.

5. The method for detecting small targets based on polarization consistency local contrast according to claim 4, characterized in that: In step (5), the adaptive threshold Defined as: ； in, and are the mean and variance of the polarization saliency map, is an adjustable constant parameter. Value greater than threshold When , the pixel is the target, otherwise it is the background.

6. A small target detection system based on polarization consistency local contrast, characterized in that: The invention comprises a memory and one or more processors, wherein the memory stores executable codes, and when the one or more processors execute the executable codes, the method for detecting small and dim targets according to any one of claims 1 to 5 is implemented.

Images