Ship target detection method in radar images considering both global and local information perception

Abstract

The present invention discloses a radar image ship target detection method that takes into account both global and local information perception. The method comprises the following steps: S1, acquiring radar image samples; S2, identifying ocean and land areas in the radar image using an adaptive threshold method; S3, extracting stable extrema regions in the image using a maximum stable extrema region method as candidate ship regions; S4, performing superpixel segmentation on the entire image using a simple linear iterative clustering method; and S5, calculating saliency values ​​for all candidate ship regions. When the saliency values ​​meet a preset condition, the candidate ship region is determined to be the final correctly detected region. The present invention comprehensively considers both global and local information perception in radar images. On the one hand, it incorporates contextual information from the scene into ship detection in complex scenes, providing more effective information. On the other hand, it combines global regional stability with local background information to accurately distinguish between real ship targets and false alarms.

Description

Technical Field

[0001] The present invention relates to the field of radar target detection, and in particular to a radar image ship target detection method that takes both global and local information perception into consideration. Background Art

[0002] While research on ship detection in synthetic aperture radar (SAR) imagery has achieved promising results in the open ocean, there is still significant room for improvement in complex scenes. Unlike the open ocean, complex scenes often feature strong land background clutter. Man-made structures such as trees, vehicles, and buildings appear similar to ships in radar images, significantly interfering with ship detection. Therefore, detecting ships in complex scenes is a challenging task.

[0003] Constant False Alarm Rate (CFAR) detection methods have been widely used for ship detection. They rely on accurate statistical modeling of background clutter to ensure optimal detection. These methods have demonstrated good performance for ship detection in simple ocean areas. However, in complex scenarios, due to strong land background clutter, existing statistical models are inapplicable and accurate modeling is difficult to achieve. Therefore, these methods are not suitable for ship detection in complex scenarios.

[0004] Currently, there are two main approaches to ship target detection in complex scenes. The first is based on land-sea segmentation. Before detection, land-sea segmentation is performed to mask land areas. Consequently, ship detection in complex scenes becomes simply sea-based ship detection. This approach significantly reduces computational complexity and prevents land interference in ship detection. However, the primary task of this approach is to achieve high-precision land-sea segmentation. Inaccurate land-sea segmentation can lead to two problems: missing land areas can still cause false alarms, and ships docked or near the shore may be misclassified as land areas during the land-sea segmentation phase, resulting in missed detections. Therefore, ship detection in complex scenes cannot be simply equated with ship detection in simple ocean areas. The second type of ship detection method in complex scenes is based on ship features, focusing on extracting more detailed and effective ship features to effectively distinguish false alarms from real ship targets. These methods improve detection performance. However, these methods also have some limitations. First, due to the extraction of more detailed ship features, these methods are often limited by computational efficiency and cannot meet real-time requirements. On the other hand, in some low- or medium-resolution scenarios, the ship's texture and outline features are less distinct, and typical ship characteristics are lost. In such cases, this method fails to extract effective ship features and is easily confused with other strongly scattering objects, making ship target identification even more difficult. Therefore, based on the above analysis, a more effective new approach is needed for ship detection in complex scenarios. Summary of the Invention

[0005] The present invention aims to overcome the shortcomings of existing technologies by providing a method for ship detection in complex scenes that comprehensively considers both global and local information perception in radar images. This method incorporates contextual information from the scene into ship detection in complex scenes, providing more effective information. Furthermore, it combines global regional stability with local background information to accurately distinguish between real ship targets and false alarms. This method improves ship detection performance and reduces false alarms in land areas.

[0006] The object of the present invention is achieved through the following technical solution: a radar image ship target detection method that takes into account both global and local information perception, comprising the following steps:

[0007] S1. Obtain radar image samples: Obtain nearshore scene radar image samples from public spaceborne synthetic aperture radar image data;

[0008] S2. Perform land-sea segmentation: Use an adaptive threshold method to identify the ocean and land areas in the radar image, and retain the buffer area where the ocean area and the land-sea boundary meet;

[0009] S3. Perform global information extraction: Use the maximum stable extreme value region method to extract the stable extreme value regions in the area obtained in S2, and use these selected extreme value regions as candidate ship regions;

[0010] S4, perform local information extraction: use a simple linear iterative clustering method to perform superpixel segmentation on the image obtained in S2;

[0011] S5. Saliency value calculation and final detection: Saliency values ​​are calculated for all candidate ship regions. When the saliency values ​​meet the preset conditions, the candidate ship regions are determined to be the final correct detection regions.

[0012] The specific implementation method of step S2 is as follows: let the total number of pixels in the radar image be N, the number of gray levels be K, and the gray histogram be H(i), where i = 0, 1, ..., K-1; H(i) represents the number of pixels with gray level i; let the threshold be t, and pixels with values ​​greater than t be considered foreground, and pixels with values ​​less than or equal to t be considered background; the pixel weights and average gray values ​​of the background and foreground are calculated as follows:

[0013]

[0014] ω2(t)=1-ω1(t),

[0015] ω1(t) and ω2(t) represent the pixel weights of background and foreground, respectively; μ1(t) and μ2(t) represent the average grayscale values ​​of background and foreground, respectively; inter-class variance Calculated by the following formula:

[0016]

[0017] Let t=0,1,…,K-1, we get choose The threshold t corresponding to the maximum value is taken as the optimal threshold t * , using the optimal threshold t * Binarize the original image so that the pixel value is greater than t * is considered as the foreground, i.e. the land area; less than t * The ocean area is regarded as the background, that is, the ocean area; after obtaining the ocean area and the land area, the area within a certain range on both sides of the dividing line between the ocean area and the land area is used as the buffer area.

[0018] The specific implementation method of step S3 is: let Q1,…,Q i-1 ,Q i is a series of nested extreme regions that satisfy Q i+1 ∈Q i , for a specific i, when the following formula holds, Q i Considered as a maximally stable extremum region:

[0019]

[0020] Among them, S(Q i ) is the region Q i represents the area of ​​the two regions, |·| represents the absolute value, Δ is the step size of the area change, and T is the maximum area change allowed between the two regions.

[0021] The specific implementation method of step S5 is as follows: Assume that a total of M candidate ship regions are obtained, denoted as m=1,…,M, and measure the significance S(Q m ) is calculated as:

[0022]

[0023] Among them, Ω Qm For region Q m The local area is defined as: m The center of the rectangle is centered on the center of the rectangle and has a side of 3w*3h; spn(Q m ) is the region Q m The number of superpixels, w(Q m ) is the weight coefficient, ds(Q m ,Q n) is the region Q m and region Q n The center point is in the local area Ω Qm The normalized spatial distance in , dg(Q m ,Q n ) is the region Q m and region Q n The center point is in the local area Ω Qm The normalized grayscale distance in , pn(Q m ) is the region Q m In the local area Ω Qm The number of normalized pixels in ;

[0024] After the saliency of each candidate region is calculated, it is compared with the pre-set threshold, and the region whose saliency exceeds this threshold is used as the final detection region.

[0025] The present invention provides a method for ship detection in complex scenarios that comprehensively considers both global and local information perception in radar images. This method incorporates contextual information from the scene into ship detection in complex scenarios, providing more effective information. Furthermore, it combines global regional stability with local background information to accurately distinguish between real ship targets and false alarms. This method improves ship detection performance and reduces false alarms in land areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Flowchart of the present invention;

[0027] Figure 2 This is a diagram showing the experimental results of radar image scene 1 in the present invention;

[0028] Figure 3 This is the experimental result diagram of radar image scene 2 in the present invention. DETAILED DESCRIPTION

[0029] The technical solution of the present invention is further described below with reference to the accompanying drawings.

[0030] like Figure 1 As shown, a radar image ship target detection method taking into account both global and local information perception of the present invention includes the following steps:

[0031] S1. Obtain radar image samples: Nearshore radar image samples are obtained from Sentinel-1 and other publicly available spaceborne synthetic aperture radar image data, freely available online from the European Space Agency (ESA). In this example, the samples are two typical complex ocean scenes. Each scene is approximately 2000 × 2000 pixels in size, with each pixel representing an actual scene size of 3 m x 3 m. The scenes are located near the Port of Singapore. The two scenes cover a mixed land and sea area of ​​approximately 170 square kilometers and include approximately 80 vessel targets.

[0032] S2. Perform land-sea segmentation: Use the adaptive thresholding method (OTSU) to identify the ocean and land areas in the radar image, and retain the buffer area where the ocean and land-sea boundaries meet. The specific implementation method is as follows: let the total number of pixels in the radar image be N, the number of gray levels be K, and the gray histogram be H(i), where i = 0, 1, ..., K-1; H(i) represents the number of pixels with gray level i; let the threshold be t, and pixels with values ​​greater than t be considered foreground, and pixels with values ​​less than or equal to t be considered background; the pixel weights and average gray values ​​of the background and foreground are calculated as follows:

[0033]

[0034] ω2(t)=1-ω1(t),

[0035] ω1(t) and ω2(t) represent the pixel weights of background and foreground, respectively; μ1(t) and μ2(t) represent the average grayscale values ​​of background and foreground, respectively; inter-class variance Calculated by the following formula:

[0036]

[0037] Let t=0,1,…,K-1, we get choose The threshold t corresponding to the maximum value is taken as the optimal threshold t * , using the optimal threshold t * Binarize the original image so that the pixel value is greater than t * is considered as the foreground, i.e. the land area; less than t * is regarded as the background, that is, the ocean area; after obtaining the ocean area and the land area, the area within a certain range on both sides of the dividing line between the ocean area and the land area is used as the buffer area, and the specific range can be selected by the user.

[0038] S3, perform global information extraction: use the maximum stable extreme region method (MSER) ​​to extract the stable extreme regions in the area obtained in S2, and use these selected extreme regions as candidate ship regions; the specific implementation method is: let Q1,…,Q i-1 ,Q i is a series of nested extreme regions that satisfy Q i+1 ∈Q i , for a specific i, when the following formula holds, Q i Considered as a maximally stable extremum region:

[0039]

[0040] Among them, S(Q i ) is the region Q i represents the area of ​​the two regions, |·| represents the absolute value, Δ is the step size of the area change, and T is the maximum area change allowed between the two regions.

[0041] S4. Perform local information extraction: Use the simple linear iterative clustering method (SLIC) to perform superpixel segmentation on the image obtained in S2. In SLIC, the method for calculating the distance between two pixels is:

[0042]

[0043] d g =|p j -p i |

[0044]

[0045] Among them, d g Represents the grayscale distance, d s Represents the spatial distance, N g and N s are the maximum grayscale distance and the maximum spatial distance within the class respectively; p i ,x i ,y i Represent the grayscale value of pixel i, the horizontal coordinate and the vertical coordinate of pixel i respectively. Each pixel calculates the distance D from the surrounding cluster centers and is then classified into the cluster center closest to it.

[0046] S5. Calculation of saliency value and final detection: In steps S3 and S4, the maximum stable region and superpixel segmentation results are obtained respectively. The saliency value of all candidate ship regions is calculated. When the saliency value meets the preset conditions, the candidate ship region is determined as the final correct detection region. The specific implementation method is as follows: Assuming that a total of M candidate ship regions are obtained, denoted as m=1,…,M, the saliency S(Q m) is calculated as:

[0047]

[0048]

[0049] in, For region Q m The local area is defined as: m The center of the rectangle is centered on the center of the rectangle and has a side of 3w*3h; spn(Q m ) is the region Q m The number of superpixels, w(Q m ) is the weight coefficient, ds(Q m ,Q n ) is the region Q m and region Q n The center point is in the local area The normalized spatial distance in , dg(Q m ,Q n ) is the region Q m and region Q n The center point is in the local area The normalized grayscale distance in , pn(Q m ) is the region Q m In local areas The number of normalized pixels in ;

[0050] After the saliency of each candidate region is calculated, it is compared with the pre-set threshold, and the region whose saliency exceeds this threshold is used as the final detection region.

[0051] Figure 2 This is the experimental result diagram of radar image scene 1 in the present invention. Figure 3 This is the experimental result diagram of radar image scene 2 in the present invention. Figure 2 and Figure 3 The square box represents a correct detection, the triangle represents a false alarm, and the circle represents a missed detection. Figure 2 and Figure 3 It can be seen that the vast majority of ship targets are detected correctly, with a high detection accuracy.

[0052] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A radar image ship target detection method that takes into account both global and local information perception is characterized by: The following steps are involved: S1. Obtain radar image samples: Obtain nearshore scene radar image samples from public spaceborne synthetic aperture radar image data; S2. Perform land-sea segmentation: Use an adaptive threshold method to identify the ocean and land areas in the radar image, and retain the buffer area where the ocean area and the land-sea boundary meet; S3. Perform global information extraction: Use the maximum stable extreme value region method to extract the stable extreme value regions in the area obtained in S2, and use these selected extreme value regions as candidate ship regions; S4, perform local information extraction: use a simple linear iterative clustering method to perform superpixel segmentation on the image obtained in S2; S5. Calculation of saliency value and final detection: Calculate the saliency value of all candidate ship regions. When the saliency value meets the preset conditions, the candidate ship region is determined to be the final correct detection region. The specific implementation method is as follows: Assuming that a total of M candidate ship regions are obtained, denoted as m=1,…,M, the saliency S(Q m ) is calculated as: in, For region Q m The local area is defined as: m The center of the rectangle is centered on the center of the rectangle and has a side of 3w*3h; spn(Q m ) is the region Q m The number of superpixels, w(Q m ) is the weight coefficient, ds(Q m ,Q n ) is the region Q m and region Q n The center point is in the local area The normalized spatial distance in , dg(Q m ,Q n ) is the region Q m and region Q n The center point is in the local area The normalized grayscale distance in , pn(Q m ) is the region Q m In local areas The number of normalized pixels in ; After the saliency of each candidate region is calculated, it is compared with the pre-set threshold, and the region whose saliency exceeds this threshold is used as the final detection region.

2. The radar image ship target detection method taking into account both global and local information perception according to claim 1 is characterized in that: The specific implementation method of step S2 is as follows: let the total number of pixels in the radar image be N, the number of gray levels be K, and the gray histogram be H(i), where i=0, 1, ..., K-1; H(i) represents the number of pixels with gray level i. Let the threshold be t, and the pixels with values ​​greater than t are considered as foreground, and those less than or equal to t are considered as background. The pixel weights and average gray values ​​of background and foreground are calculated as follows: ω1(t) and ω2(t) represent the pixel weights of background and foreground, μ1(t) and μ2(t) represent the average grayscale values ​​of background and foreground, respectively; inter-class variance Calculated by the following formula: Let t=0,1,…,K-1, we get choose The threshold t corresponding to the maximum value is taken as the optimal threshold t * , using the optimal threshold t * Binarize the original image so that the pixel value is greater than t * is considered as the foreground, i.e. the land area; less than t * The ocean area is regarded as the background, that is, the ocean area; after obtaining the ocean area and the land area, the area within a certain range on both sides of the dividing line between the ocean area and the land area is used as the buffer area.

3. The radar image ship target detection method taking into account both global and local information perception according to claim 1 is characterized in that: The specific implementation method of step S3 is: let Q1, ..., Q i-1 ,Q i is a series of nested extreme regions that satisfy Q i+1 ∈Q i , for a specific i, when the following formula holds, Q i Considered as a maximally stable extremum region: Among them, S(Q i ) is the region Q i represents the area of ​​the two regions, |·| represents the absolute value, Δ is the step size of the area change, and T is the maximum area change allowed between the two regions.

Images