A ship target detection method based on infrared polarization imaging

Abstract

The invention discloses a ship target detection method by infrared polarization imaging, which relates to the technical field of ship target detection and comprises the following steps: S1: acquiring infrared intensity images of three polarization angles by an infrared polarization imaging system, and calculating and acquiring a Stokes parameter image; S2: calculating and acquiring a low-noise infrared polarization degree image by representing the Stokes parameter image by a potential low-rank representation; S3: decomposing, denoising and fusing the Stokes parameter image by a potential low-rank representation, and acquiring a fused infrared polarization image; S4: acquiring a significant area image of the fused infrared polarization image by a transition area segmentation method; S5: reconstructing the low-noise infrared polarization degree image and the fused infrared polarization image based on the significant area image, and acquiring an enhanced infrared polarization image; under low-contrast conditions, utilizing the difference in infrared polarization characteristics of ship targets and backgrounds, reconstructing the polarization characteristics of weak targets, suppressing background noise interference, and enhancing the detection capability of ship targets.

Description

Technical Field

[0001] The invention relates to the technical field of ship target detection, and in particular to a ship target detection method using infrared polarization imaging. Background Art

[0002] Infrared intensity imaging relies on the radiation difference between the target and the background to perform imaging, and can complete tasks such as detection, identification, and tracking of targets during the day and night. During the infrared detection of ship targets, the infrared radiation of the ship, the sea surface, and the sky background changes continuously due to the dynamic changes in solar radiation and atmospheric thermal convection. When the radiance difference between the ship and the background is less than the minimum detection sensitivity of the infrared detection system, the contrast between the target and the background in the intensity image is low, making it difficult for the infrared detection system to effectively detect the target, which poses a serious challenge to the detection of ship targets.

[0003] As a new detection technology, infrared polarization imaging can simultaneously obtain the infrared intensity and polarization information of the target. Infrared polarization imaging is based on the difference in infrared polarization characteristics between the target and the background. It can effectively suppress the background and highlight the edge and detail features of the target, providing a new method and approach for infrared detection of low-contrast ship targets.

[0004] However, in the prior art, the infrared polarization characteristics of the ship target and background obtained under low contrast are weak, and the acquisition of the polarization image requires nonlinear calculation, which makes the background noise of the polarization image large, resulting in low contrast between the ship target and the background, affecting the detection performance of the ship target.

[0005] Therefore, a ship target detection method based on infrared polarization imaging is provided to solve the above problems. Summary of the invention

[0006] The purpose of the present invention is to provide a ship target detection method based on infrared polarization imaging. Under low contrast conditions, the polarization characteristics of the target are reconstructed by utilizing the difference between the polarization characteristics of the ship target and the background, the background noise interference is suppressed, the contrast and signal-to-noise ratio of the target and the background are improved, and the detection capability of the ship target is enhanced.

[0007] To achieve the above object, the present invention provides a ship target detection method using infrared polarization imaging, comprising the following steps:

[0008] S1: Acquire infrared intensity images at three polarization angles through infrared polarization imaging system I 0° , I 60° and I 120° , the Stokes vector is calculated S , get the Stokes vector S Parameters S0. S 1 and S 2. Construct the Stokes image, which includes parameters S 0 images, parameters S 1 Images and parameters S 2 images;

[0009] S2: LatLRR parameters via latent low-rank representation S 1 Images and parameters S 2 images are represented by low rank and the parameters are obtained S 1 Image base layer and parameters S 2 image base layers, combined with parameters S 0 image to obtain low-noise infrared polarization images LRDoLP ;

[0010] S3: LatLRR decomposition parameters via latent low-rank representation S 1 Images and parameters S 2 images, the decomposed parameters S 1 Images and parameters S 2 Image fusion, obtaining fused infrared polarization image REDoLP ;

[0011] S4: Using fused infrared polarization images REDoLP The local complexity and local variance of the transition region description operator are constructed NLCV , using the transition region segmentation method to obtain the fused infrared polarization image REDoLP The salient area image ;

[0012] S5: Based on salient area image , the low-noise infrared polarization image LRDoLP and fused infrared polarization images REDoLP Reconstruct and obtain enhanced infrared polarization image ENDoLP .

[0013] Preferably, in step S1, the Stokes vector S It is expressed as:

[0014] ;

[0015] in, S 0 represents the total intensity of infrared radiation, S 1 represents the intensity difference between the horizontal and vertical linear polarized light. S 2 represents the intensity difference between two diagonally polarized lights.

[0016] Preferably, step S2 specifically includes the following steps:

[0017] S21: Based on the low-rank characteristics of the Stokes image, the Stokes parameters are represented by low-rank approximation S 1 and S 2. Remove the noise of Stokes image;

[0018] S22: Matrix recovery is performed through latent low-rank representation LatLRR, hidden items are added to the dictionary, and noise of the Stokes image is suppressed;

[0019] S23: Training the decomposition matrix containing significant coefficients L , respectively for the parameters S 1 Images and parameters S 2. Perform LatLRR decomposition on the image to obtain the parameters S 1 Image base layer, parameters S 1 Image saliency layer, parameters S 2 Image base layer and parameters S 2 Image saliency layer;

[0020] S24: By parameters S 1 Image base layer, parameters S 2 Image base layer and parameters S 0 Image calculation low noise infrared polarization image LRDoLP .

[0021] Preferably, step S23 specifically includes the following steps:

[0022] Step 1: Use sliding window to set parameters S 1 Images and parameters S 2. The image is divided into several sub-blocks;

[0023] Step 2: Set the parameters S 1 Image sub-blocks and parameters S 2 The image sub-blocks are rearranged into columns to obtain a matrix image , ,exist In , each column represents an image block;

[0024] Step 3: By decomposing the matrix L For matrix images Decompose and obtain the decomposed saliency matrix ;

[0025] Step 4: Reconstruct the saliency matrix , get the parameters S 1 Image saliency layer and parameters S 2 Image saliency layer, saliency layer It is expressed as:

[0026] ;

[0027] in, R Represents the saliency matrix Reconstructed into salient layers Operations of

[0028] Step 5: Parameters S 1 Images and parameters S 2 images are subtracted from the parameters S 1 Images and parameters S 2 The significant layer in the image, get the parameters S 1 Image base layer and parameters S 2 Image base layer, base layer It is expressed as:

[0029] .

[0030] Preferably, in step S24, the low-noise infrared polarization image LRDoLP It is expressed as:

[0031] ;

[0032] in, Representation parameters S 1 image base layer, Representation parameters S 2 Image base layer.

[0033] Preferably, step S3 specifically includes the following steps:

[0034] S31: Fusion of base layers via weighted averaging strategy based on visual saliency map (VSM) , get the base layer The fused image , base layer The fused image It is expressed as:

[0035] ;

[0036] ;

[0037] in, W b represents the weight parameter, Representation parameters S 1 Visual Saliency Map VSM of the image, Representation parameters S 2 Visual saliency map VSM of the image;

[0038] S32: Guided filtering method for the salient layer Perform noise reduction to suppress the significant layer Noise, highlighting the salient features of the target edge and texture. The window size of the guided filter is set to 8 and the regularization parameter is set to 0.2;

[0039] S33: The fusion weight coefficient is obtained by taking the maximum fusion strategy of the absolute value. The fusion weight coefficient is expressed as:

[0040] ;

[0041] in, j Represents the salient layer Pixels;

[0042] The parameters S 1 Image saliency layer and parameters S 2 Image saliency layer fusion to obtain saliency layer The fused image P SF , significant layer The fused image P SF It is expressed as:

[0043] ;

[0044] in, Representation parameters S 1 image saliency layer, Representation parameters S 2 Image saliency layer;

[0045] S34: Base layer The fused image and salient layer The fused image P SF Superposition and summation to obtain a fused infrared polarization image REDoLP , fused infrared polarization image REDoLP It is expressed as:

[0046] .

[0047] Preferably, step S4 specifically includes the following steps:

[0048] S41: For fusion of infrared polarization images REDoLP Pixels , calculate the local complexity Lc and variance , local complexity Lc and variance Respectively expressed as:

[0049] ;

[0050] ;

[0051] ;

[0052] in, Represents pixels Corresponding local area, local area The size is set to , The size is set to 3, Represents a local area The average gray value of Represents pixels The corresponding gray value;

[0053] S42: Local complexity Lc and variance Normalize it and get the normalized local complexity NLc and the normalized variance , the normalized local complexity NLc and the normalized variance Synthesize into transition region description operator NLCV , transition region description operator NLCV It is expressed as:

[0054] ;

[0055] in, ω Represents the weight coefficient, weight coefficient ω Set to 0.2;

[0056] S43: Transition region description operator NLCV Convert to vector SNLCV , the vector SNLCV Sort all elements of in descending order and select the vector SNLCV Before The pixels corresponding to the elements are used as transition pixels to construct the transition area. α is a parameter constant. α Set to 0.1, is the total number of pixels;

[0057] S44: Extract the real significant transition area in the transition area, remove the false transition area contained in the object and background through morphological opening operation, and obtain the significant area image , salient region image It is expressed as:

[0058] .

[0059] Preferably, in step S44, the structure element of the morphological opening operation is set to a square structure element, and the width of the square structure element is set to 2 pixels.

[0060] Preferably, step S5 specifically includes the following steps:

[0061] S51: Low-noise infrared polarization image LRDoLP , fusion of infrared polarization images REDoLP and salient region image Weighted summation to obtain enhanced infrared polarization image ENDoLP , Enhanced infrared polarization images ENDoLP It is expressed as:

[0062] ;

[0063] S52: Enhancement of infrared polarization images by smoothing filtering ENDoLP Processing to soften and enhance infrared polarization images ENDoLP At the jagged edges of the salient areas, the filter mask of the smoothing filter is set to [1 / 32,1 / 32,1 / 32;1 / 32,1 / 32,1 / 32;1 / 32,1 / 32,1 / 32].

[0064] Therefore, the present invention adopts the above-mentioned ship target detection method of infrared polarization imaging, which has the following beneficial effects:

[0065] (1) Infrared polarization imaging technology can simultaneously obtain the infrared intensity and polarization information of the target. Infrared polarization imaging is based on the difference in infrared polarization characteristics between the target and the background. It can effectively suppress the background and highlight the edge and detail features of the target.

[0066] (2) Under low-contrast conditions, the polarization characteristics of the target are reconstructed by utilizing the difference between the polarization characteristics of the ship target and the background, which suppresses the interference of background noise, improves the contrast and signal-to-noise ratio between the target and the background, and significantly enhances the infrared polarization detection capability of the ship target.

[0067] The method scheme of the present invention is further described in detail below through the drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is a flow chart of a ship target detection method using infrared polarization imaging according to the present invention;

[0069] Figure 2 Infrared intensity images at three polarization angles according to the embodiment of the present invention I 0° , I 60° and I 120° ;

[0070] Figure 3 The infrared intensity of the embodiment of the present invention S 0 images;

[0071] Figure 4 This is the infrared polarization image of the embodiment of the present invention DoLP ;

[0072] Figure 5 This is a low-noise infrared polarization image according to an embodiment of the present invention. LRDoLP ;

[0073] Figure 6 Fusion of infrared polarization images for embodiments of the present invention REDoLP ;

[0074] Figure 7 is a salient area image according to an embodiment of the present invention mask ;

[0075] Figure 8 Enhanced infrared polarization image for embodiments of the present invention ENDoLP ;

[0076] Figure 9 Schematic diagram of the contrast of the experimental results of the embodiment of the present invention, where (a) the infrared intensity S 0 image, (b) linear polarization image DoLP , (c) infrared polarization image equalization enhancement DPHE method, (d) multi-level Gaussian curvature filtering image decomposition and fusion MLGCF method of infrared intensity image and linear polarization image, (e) multi-layer latent low rank representation decomposition and fusion MDLatLRR method of infrared intensity and linear polarization image, (f) the present invention. DETAILED DESCRIPTION

[0077] The method scheme of the present invention is further described below through drawings and embodiments.

[0078] Unless otherwise defined, method terms or scientific terms used in the present invention shall have the common meanings understood by one of ordinary skill in the art to which the present invention belongs.

[0079] The words "include" or "comprises" and the like used in the present invention mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of also including other elements. The orientation or position relationship indicated by the terms "inside", "outside", "upper", "lower", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention. When the absolute position of the described object changes, the relative position relationship may also change accordingly. In the present invention, unless otherwise clearly specified and limited, the terms "attachment" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral body; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0080] Example

[0081] like Figures 1 - 9 As shown, the present invention provides a ship target detection method using infrared polarization imaging, comprising the following steps:

[0082] S1: Acquire infrared intensity images at three polarization angles through infrared polarization imaging system I 0° , I 60° and I 120° , the Stokes vector is calculated S , get the Stokes vector S Parameters S 0. S 1 and S 2. Construct the Stokes image, which includes parameters S 0 images, parameters S 1 Images and parameters S 2 images;

[0083] In step S1, the Stokes vector S It is expressed as:

[0084] ;

[0085] in, S 0 represents the total intensity of infrared radiation, S 1 represents the intensity difference between the horizontal and vertical linear polarized light. S 2 represents the intensity difference between two diagonally polarized lights.

[0086] S2: There are inherent structural similarities between different parts of the Stokes parameter image. The eigenvalues ​​of the image matrix are mainly distributed on smaller values, and the values ​​of most singular values ​​are low, indicating that the characteristic information of the image data only occupies part of the content, and the image has obvious low-rank characteristics;

[0087] Therefore, the LatLRR parameter is S 1 Images and parameters S 2 images are represented by low rank and the parameters are obtained S 1 Image base layer and parameters S 2 image base layers, combined with parameters S 0 image to obtain low-noise infrared polarization images LRDoLP ;

[0088] Step S2 specifically includes the following steps:

[0089] S21: Based on the low-rank characteristics of the Stokes image, the Stokes parameters are represented by low-rank approximation S 1 and S 2. Remove the noise of Stokes image;

[0090] S22: Matrix recovery is performed through latent low-rank representation LatLRR, hidden items are added to the dictionary, and noise of the Stokes image is suppressed;

[0091] S23: Training the decomposition matrix containing significant coefficients L , respectively for the parameters S 1 Images and parameters S 2. Perform LatLRR decomposition on the image to obtain the parameters S 1 Image base layer, parameters S 1 Image saliency layer, parameters S 2 Image base layer and parameters S 2 Image saliency layer;

[0092] Step S23 specifically includes the following steps:

[0093] Step 1: Use sliding window to set parameters S 1 Images and parameters S 2. The image is divided into several sub-blocks;

[0094] Step 2: Set the parameters S 1 Image sub-blocks and parameters S 2 The image sub-blocks are rearranged into columns to obtain a matrix image , ,exist In , each column represents an image block;

[0095] Step 3: By decomposing the matrix LFor matrix images Decompose and obtain the decomposed saliency matrix ;

[0096] Step 4: Reconstruct the saliency matrix , get the parameters S 1 Image saliency layer and parameters S 2 Image saliency layer, saliency layer It is expressed as:

[0097] ;

[0098] in, R Represents the saliency matrix Reconstructed into salient layers Operations of

[0099] Step 5: Remove parameters separately S 1 Images and parameters S 2 The significant layer in the image, get the parameters S 1 Image base layer and parameters S 2 Image base layer, base layer It is expressed as:

[0100] ;

[0101] S24: Parameters S 1 Images and parameters S 2 The image is obtained through nonlinear operation, which makes the noise transmission and amplification, linear polarization DoLP It is directly calculated from the Stokes parameters, which makes the polarization image sensitive to noise and affects the quality of the polarization image;

[0102] Linear polarization image DoLP It is the basic characteristic of infrared polarization, linear polarization image DoLP It is expressed as:

[0103] .

[0104] So by parameter S 1 Image base layer, parameters S 2 Image base layers and parameters S 0 Image calculation low noise infrared polarization image LRDoLP ,parameter S 0 image is intensity image;

[0105] In step S24, the low-noise infrared polarization image LRDoLP It is expressed as:

[0106] ;

[0107] in, Representation parameters S 1 image base layer, Representation parameters S 2 Image base layer.

[0108] S3: In polarization vector space, the parameters S 1 Images and parameters S 2 The image contains target detail feature information at different levels. In order to obtain an image with clear edges, prominent contours, and rich details, the LatLRR decomposition parameters are represented by the potential low rank S 1 Images and parameters S 2 images, the decomposed parameters S 1 Images and parameters S 2 Image fusion, obtaining fused infrared polarization image REDoLP , avoiding nonlinear computational amplification of image noise;

[0109] Step S3 specifically includes the following steps:

[0110] S31: The base layer contains the structural information and contrast information of the source image. When images are fused, this information needs to be retained to the greatest extent. In practical applications, due to the limited number of decomposition layers, a considerable amount of residual low-frequency information is retained in the base layer. However, the average weighted fusion rule cannot effectively integrate the low-frequency information of the image, which often leads to a decrease in the contrast of the base layer fused image.

[0111] Therefore, the base layer is fused by a weighted average strategy based on the visual saliency map VSM , get the base layer The fused image , base layer The fused image It is expressed as:

[0112] ;

[0113] ;

[0114] in, W b represents the weight parameter, Representation parameters S 1 Visual Saliency Map VSM of the image, Representation parameters S 2 Visual saliency map VSM of the image;

[0115] S32: Due to the nonlinear transmission and amplification of image noise, the Stokes parameter image is inevitably affected by noise. The noise in the Stokes image is similar to the significant feature information. S 1 Images and parametersS 2. Potential low-rank decomposition of the image. The noise is decomposed into the significant layer of the image as significant information. The existence of noise seriously affects the quality of the polarization fusion image.

[0116] Therefore, the salient layer is Perform noise reduction to suppress the significant layer Noise, highlighting the salient features of the target edge and texture. The window size of the guided filter is set to 8 and the regularization parameter is set to 0.2;

[0117] S33: The salient layer image contains the detailed texture and structural information of the image. The detailed features of the fused image are enhanced by superimposing and summing the salient layer images. However, the salient layer image also contains the noise information in the image. The superimposing and summing strategy increases the noise of the polarization image and affects the image quality.

[0118] Therefore, the fusion weight coefficient is obtained by taking the maximum fusion strategy of absolute value. The fusion weight coefficient is expressed as:

[0119] ;

[0120] in, j Represents the salient layer Pixels;

[0121] The parameters S 1 Image saliency layer and parameters S 2 Image saliency layer fusion to obtain saliency layer The fused image P SF , significant layer The fused image P SF It is expressed as:

[0122] ;

[0123] in, Representation parameters S 1 image saliency layer, Representation parameters S 2 Image saliency layer;

[0124] S34: Base layer The fused image and salient layer The fused image P SF Superposition and summation to obtain a fused infrared polarization image REDoLP , fused infrared polarization image REDoLP It is expressed as:

[0125] .

[0126] S4: Using fused infrared polarization images REDoLP The local complexity and local variance of the transition region description operator are constructed NLCV , using the transition region segmentation method to obtain the fused infrared polarization image REDoLP The salient area image ;

[0127] Step S4 specifically includes the following steps:

[0128] S41: The transition area of ​​the infrared polarization image has structural information similar to the edge, exists near the edge of the image, has a certain width, and is used for fusion of infrared polarization images. REDoLP Pixels , calculate the local complexity Lc and variance , local complexity Lc and variance Respectively expressed as:

[0129] ;

[0130] ;

[0131] ;

[0132] in, Represents pixels Corresponding local area, local area The size is set to , The size is set to 3, Represents a local area The average gray value of Represents pixels The corresponding gray value;

[0133] S42: Local complexity Lc and variance Normalization is performed to prevent one of the image matrices from being ignored due to its large value difference, and the normalized local complexity is obtained. NLc and the normalized variance , the normalized local complexity NLc and the normalized variance Synthesized into transition region description operator NLCV , transition region description operator NLCV It is expressed as:

[0134] ;

[0135] in, ωRepresents the weight coefficient, weight coefficient ω Set to 0.2;

[0136] S43: Transition region description operator NLCV Convert to vector SNLCV , the vector SNLCV Sort all elements of in descending order and select the vector SNLCV Before The pixels corresponding to the elements are used as transition pixels to construct the transition area. α is a parameter constant. α Set to 0.1, is the total number of pixels;

[0137] S44: Since the grayscale changes in the local neighborhood are frequent and dense, the transition pixels should have greater local complexity and local variance than the non-transition pixels. This feature can be used to extract the transition area, extract the real significant transition area in the transition area, and remove the false transition area contained in the object and background through morphological opening operation to obtain the significant area image. mask , salient region image mask It is expressed as:

[0138] ;

[0139] In step S44, the structure element of the morphological opening operation is set to a square structure element, and the width of the square structure element is set to 2 pixels.

[0140] S5: In order to effectively suppress background noise and highlight the characteristic area of ​​the target, based on the salient area image , the low-noise infrared polarization image LRDoLP and fused infrared polarization images REDoLP Reconstruct and obtain enhanced infrared polarization image ENDoLP ;

[0141] Step S5 specifically includes the following steps:

[0142] S51: Low-noise infrared polarization image LRDoLP , fusion of infrared polarization images REDoLP and salient region image Weighted summation to obtain enhanced infrared polarization image ENDoLP , Enhanced infrared polarization images ENDoLP It is expressed as:

[0143] ;

[0144] S52: Since the reconstructed infrared polarization image is obtained by extracting the significant feature area of ​​the image and weighting the polarization degree image represented by the low rank, the image has jagged edges in the significant area;

[0145] Enhancement of infrared polarization images by smoothing filtering ENDoLP Processing to soften and enhance infrared polarization images ENDoLP Preserve image edge details on jagged edges in prominent areas, and improve the visual effect of infrared polarization enhanced images;

[0146] The filter mask for the smoothing filter is set to [1 / 32,1 / 32,1 / 32;1 / 32,1 / 32,1 / 32;1 / 32,1 / 32,1 / 32].

[0147] This embodiment is experimentally compared with the infrared polarization degree image equalization enhancement DPHE, the multi-level Gaussian curvature filtering image decomposition and fusion MLGCF of infrared intensity image and linear polarization degree image, and the multi-layer latent low-rank representation decomposition and fusion MDLatLRR of infrared intensity and linear polarization degree images;

[0148] The box is the selected target area, and the target adjacent area is the background area. The results are evaluated using local contrast and signal-to-noise ratio indicators. The evaluation indicators of the experimental results are shown in Table 1:

[0149] Table 1: Experimental results evaluation indicators

[0150] ;

[0151] As shown in Table 1, the contrast and signal-to-noise ratio of the technical solution of this embodiment are much higher than those of the infrared intensity S 0 image, linear polarization image DoLP Compared with other methods in the comparative experiments, it is shown that the technical solution of this embodiment is superior to other methods in terms of the ability to improve image contrast and signal-to-noise ratio, and the target detection ability.

[0152] Therefore, the present invention adopts the above-mentioned infrared polarization imaging ship target detection method, and utilizes the difference in polarization characteristics between ship targets and background under low contrast conditions to reconstruct the target polarization characteristics, suppress background noise interference, improve the contrast and signal-to-noise ratio between the target and the background, and enhance the detection capability of ship targets.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the method scheme of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, ordinary method personnel in the field should understand that they can still modify or replace the method scheme of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified method scheme to deviate from the spirit and scope of the method scheme of the present invention.

Claims

1. A ship target detection method using infrared polarization imaging, characterized in that: The following steps are involved: S1: Acquire infrared intensity images at three polarization angles through infrared polarization imaging system I 0° , I 60° and I 120° , the Stokes vector is calculated S , get the Stokes vector S Parameters S 0. S 1 and S 2. Construct the Stokes image, which includes parameters S 0 images, parameters S 1 Images and parameters S 2 images; S2: LatLRR parameters via latent low-rank representation S 1 Images and parameters S 2 images are represented by low rank and the parameters are obtained S 1 Image base layer and parameters S 2 image base layers, combined with parameters S 0 image to obtain low-noise infrared polarization images LRDoLP ; Step S2 specifically includes the following steps: S21: Based on the low-rank characteristics of the Stokes image, the Stokes parameters are represented by low-rank approximation S 1 and S 2. Remove the noise of Stokes image; S22: Matrix recovery is performed through latent low-rank representation LatLRR, hidden items are added to the dictionary, and noise of the Stokes image is suppressed; S23: Training the decomposition matrix containing significant coefficients L , respectively for the parameters S 1 Images and parameters S 2. Perform LatLRR decomposition on the image to obtain the parameters S 1 Image base layer, parameters S 1 Image saliency layer, parameters S 2 Image base layer and parameters S 2 Image saliency layer; S24: By parameters S 1 Image base layer, parameters S 2 Image base layer and parameters S 0 Image calculation low noise infrared polarization image LRDoLP ; S3: LatLRR decomposition parameters via latent low-rank representation S 1 Images and parameters S 2 images, the decomposed parameters S 1 Images and parameters S 2 Image fusion, obtaining fused infrared polarization image REDoLP ; S4: Using fused infrared polarization images REDoLP The local complexity and local variance of the transition region description operator are constructed NLCV , using the transition region segmentation method to obtain the fused infrared polarization image REDoLP The salient area image mask ; S5: Based on salient area image mask , the low-noise infrared polarization image LRDoLP and fused infrared polarization images REDoLP Reconstruct and obtain enhanced infrared polarization image ENDoLP .

2. The ship target detection method using infrared polarization imaging according to claim 1, characterized in that: In step S1, the Stokes vector S It is expressed as: ； in, S 0 represents the total intensity of infrared radiation, S 1 represents the intensity difference between the horizontal and vertical linear polarized light. S 2 represents the intensity difference between two diagonally polarized lights.

3. The ship target detection method using infrared polarization imaging according to claim 1, characterized in that: Step S23 specifically includes the following steps: Step 1: Use sliding window to set parameters S 1 Images and parameters S 2. The image is divided into several sub-blocks; Step 2: Set the parameters S 1 Image sub-blocks and parameters S 2 The image sub-blocks are rearranged into columns to obtain a matrix image M ( S k ), S k ∈( S 1, S 2) In M ( S k ), each column represents an image block; Step 3: By decomposing the matrix L For matrix images M ( S k ) to decompose and obtain the decomposed saliency matrix ; Step 4: Reconstruct the saliency matrix , get the parameters S 1 Image saliency layer and parameters S 2 Image saliency layer, saliency layer It is expressed as: ； in, R Represents the saliency matrix Reconstructed into salient layers Operations of Step 5: Parameters S 1 Images and parameters S 2 images are subtracted from the parameters S 1 Images and parameters S 2 The significant layer in the image, get the parameters S 1 Image base layer and parameters S 2 Image base layer, base layer It is expressed as: 。 4. The ship target detection method using infrared polarization imaging according to claim 1, characterized in that: In step S24, the low-noise infrared polarization image LRDoLP It is expressed as: ； in, Representation parameters S 1 image base layer, Representation parameters S 2 Image base layer.

5. The ship target detection method using infrared polarization imaging according to claim 1, characterized in that: Step S3 specifically includes the following steps: S31: Fusion of base layers via weighted averaging strategy based on visual saliency map (VSM) , get the base layer The fused image , base layer The fused image It is expressed as: ； ； in, W b represents the weight parameter, Representation parameters S 1 Visual Saliency Map VSM of the image, Representation parameters S 2 Visual saliency map VSM of the image; S32: Guided filtering method for the salient layer Perform noise reduction to suppress the significant layer Noise, highlighting the salient features of the target edge and texture, the window size of the guided filter is set to 8, and the regularization parameter is set to 0.2; S33: The fusion weight coefficient is obtained by taking the maximum fusion strategy of the absolute value. The fusion weight coefficient is expressed as: ； in, j Represents the salient layer Pixels; The parameters S 1 Image saliency layer and parameters S 2 Image saliency layer fusion to obtain saliency layer The fused image P SF , significant layer The fused image P SF It is expressed as: ； in, Representation parameters S 1 image saliency layer, Representation parameters S 2 Image saliency layer; S34: Base layer The fused image and salient layer The fused image P SF Superposition and summation to obtain a fused infrared polarization image REDoLP , fused infrared polarization image REDoLP It is expressed as: 。 6. The ship target detection method using infrared polarization imaging according to claim 1, characterized in that: Step S4 specifically includes the following steps: S41: For fusion of infrared polarization images REDoLP Pixels , calculate the local complexity and variance , local complexity and variance Respectively expressed as: ； ； ； in, Represents pixels Corresponding local area, local area The size is set to , m The size is set to 3, Represents a local area The average gray value of Represents pixels The corresponding gray value; S42: Local complexity and variance Normalize the local complexity and the normalized variance Synthesized into transition region description operator NLCV , transition region description operator NLCV It is expressed as: ； in, ω Represents the weight coefficient, weight coefficient ω Set to 0.2; S43: Transition region description operator NLCV Convert to vector SNLCV , the vector SNLCV Sort all elements of in descending order and select the vector SNLCV Before The pixels corresponding to the elements are used as transition pixels to construct the transition area. α is a parameter constant. α Set to 0.1, is the total number of pixels; S44: Extract the real significant transition area in the transition area, remove the false transition area contained in the object and background through morphological opening operation, and obtain the significant area image , salient region image It is expressed as: 。 7. The ship target detection method using infrared polarization imaging according to claim 6, characterized in that: In step S44, the structure element of the morphological opening operation is set to a square structure element, and the width of the square structure element is set to 2 pixels.

8. The ship target detection method using infrared polarization imaging according to claim 1, characterized in that: Step S5 specifically includes the following steps: S51: Low-noise infrared polarization image LRDoLP , fusion of infrared polarization images REDoLP and salient region image mask Weighted summation to obtain enhanced infrared polarization image ENDoLP , Enhanced infrared polarization images ENDoLP It is expressed as: ； S52: Enhancement of infrared polarization images by smoothing filtering ENDoLP Processing to soften and enhance infrared polarization images ENDoLP At the jagged edges of the salient areas, the filter mask of the smoothing filter is set to [1 / 32,1 / 32,1 / 32;1 / 32,1 / 32,1 / 32;1 / 32,1 / 32,1 / 32].

Images