A Ship Target Detection Scheme Based on Noise Reduction in Water Environment

Abstract

The present invention discloses a ship target detection solution based on water environment noise reduction, which combines two filtering methods: gradient blur filtering and a water pattern noise classifier based on a support vector machine. The size and feature richness of ship targets and water pattern noises in channel monitoring images are correlated with the position of the targets in the pictures. Therefore, a gradient filtering algorithm formula and an area size classifier based on a support vector machine are obtained. Through iterative learning, the classification boundary line of the area sizes of ship targets and noise targets is obtained, and a noise filtering formula based on a support vector machine is obtained. By combining multiple noise reduction algorithms, the water pattern noise filtering process is performed on the output result of motion target recognition. The on-site environmental test results show that all water environment noises can be filtered out, while still accurately capturing ship targets, improving the accuracy of the motion detection algorithm for ship recognition, meeting the application requirements of ship target recognition, and providing an accurate basis for ship monitoring and scheduling.

Description

Technical Field

[0001] The present invention relates to the technical field of ship target detection, and specifically provides a ship target detection solution based on underwater environment noise reduction. Background Art

[0002] With the rapid development of inland waterway transportation, the types and numbers of ships in the waterway are increasing continuously. There are not only large ships such as cargo ships and cruise ships, but also various small ships such as fishing boats, agricultural self-provided ships, and yachts, making the waterway and port more crowded, and the risk of ship collisions also increasing. How to establish an intelligent ship supervision platform to identify, locate, and visually supervise all ships in key waters is an urgent problem to be solved in current waterway safety supervision.

[0003] Traditional ship monitoring systems are mainly implemented through AIS and VITS, which are passive supervision means. Terminal devices need to be installed on ships. However, small ships such as fishing boats, agricultural self-provided ships, and water yachts do not have the conditions to install AIS and VITS, and some of the installed terminals may go offline, resulting in monitoring blind spots in existing ship monitoring systems. At the same time, AIS and VITS have the disadvantages of positioning errors and data update delays. There are obvious errors between the actual position of the ship and the position obtained in the system. The ship monitoring system based on AIS and VITS cannot meet the requirements of comprehensive, accurate, and reliable waterway ship monitoring. Summary of the Invention

[0004] The purpose of the present invention is to provide a ship target detection solution based on underwater environment noise reduction to overcome the interference of complex and continuous noise in the underwater environment on the recognition of moving target detection technology in the application scenario of ship monitoring.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A ship target detection solution based on underwater environment noise reduction combines two filtering methods: gradient fuzzy filtering and a water pattern noise classifier based on support vector machines.

[0006] The specific solution is as follows: According to the imaging characteristics of high-depth waterway monitoring, the distribution of underwater environment noise is mostly near the monitoring perspective, and at the same distance, the area of underwater environment noise is significantly different from the area of ship targets;

[0007] The size and feature richness of ship targets and water pattern noises in waterway monitoring images are correlated with the position of the target in the image. Ship targets with a large area and more water pattern noises are closer and located below the monitoring image, while ship targets with a small area and less water pattern noises are farther and located in the upper half of the monitoring image. Therefore, the gradient filtering algorithm is as shown in (1).

[0008]

[0009] Among them, h represents the ordinate of the pixel point in the image, that is, the water area filtering template near the image (closer to the lower part of the image) is larger, and the water area filtering template far away (located above the image) is smaller.

[0010] Noise filtering based on support vector machine: Some small water ripple noises can be filtered out through the gradient blur algorithm, but the water ripple noises with rich features such as waves cannot be effectively filtered out. This noise filtering method based on support vector machine statistically analyzes the area ratio and distance of a large number of ship targets and water ripple noises. There is an obvious boundary area in the spatial relationship between ships and water ripple noises. Therefore, based on the area size classifier of support vector machine, through iterative learning, the classification boundary line of the area size of ship targets and noise targets is obtained, as shown in Equation (2). x represents the distance between the moving target frame and the camera, and y represents the area ratio of the moving target frame to the picture. According to this classifier, it is identified which belong to real ship targets and which belong to water ripple noises, and the water ripple noise target boxes are filtered out to further achieve the effect of removing water ripple noises.

[0011]

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: By combining multiple noise reduction algorithms, the water ripple noise filtering process is carried out on the output result of moving target recognition. The on-site environmental test results show that all water environment noises can be filtered out, and at the same time, the accurate capture of ship targets is still achieved, improving the accuracy of the ship recognition of the motion detection algorithm, meeting the application requirements of ship target recognition, and providing an accurate basis for ship monitoring and dispatching.

[0013] 1. Gradient blur filtering algorithm: Fully considering the change in the target area generated by ships from far to near in channel monitoring, a gradient filtering template size is designed. On the basis of blurring the near image, the detailed features of small target ships in the distance are still retained.

[0014] 2. Filtering algorithm based on support vector machine: Aiming at the spatial distribution characteristics of noise and ship targets, the classification boundary between the two is obtained through iterative learning. It is an adaptive water ripple noise filtering algorithm, and the target boxes belonging to noise are classified, identified and deleted, and finally the ideal effect of removing water ripple noises is achieved. Description of the Drawings

[0015] Figure 1 It is a comparison diagram of water ripple noise and target size in the embodiment;

[0016] Figure 2 It is a schematic diagram of gradient filtering in this embodiment;

[0017] Figure 3 It is a relationship diagram of the area ratio of ship and water ripple noise targets to distance;

[0018] Figure 4 The figure shows the effect diagrams of water ripple noise reduction by two methods: gradient blur filtering and a water ripple noise classifier based on a support vector machine. Specific implementation manners

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Embodiment: A ship target detection solution based on water environment noise reduction combines two filtering methods: gradient blur filtering and a water ripple noise classifier based on a support vector machine.

[0021] According to the imaging characteristics of high-depth channel monitoring, the distribution of water environment noise is mostly in the near area of the monitoring perspective. Its size and distribution rules are as Figure 1 shown. The black border represents water ripple noise, and the white border represents the actual ship target. Water ripple noise is mostly in the near area of the monitoring field of view, that is, the lower part of the picture. At the same time, at the same distance, the area ratio of water environment noise to the ship target area has a significant difference. Based on the characteristics of water ripple noise, the following improvements in motion detection and noise reduction are made in this embodiment:

[0022] (1) Gradient blur filtering: In the channel monitoring image, the size and feature richness of the ship target and water ripple noise are correlated with the position of the target in the picture. The ship target with a short distance and in the lower part of the image has a larger area and more water ripple noise, while the ship target with a long distance and in the upper half of the monitoring image has a smaller area and less water ripple noise. Therefore, it is impossible to perform blur filtering on the entire picture with a unified filtering template. In this application, a gradient filtering algorithm is designed according to the target distribution characteristics of channel monitoring. The size of its filtering template is as shown in formula (1), where h represents the ordinate of the pixel point in the image, that is, the filtering template for the water area closer to the lower part of the image is larger, and the filtering template for the water area located above the image in the distance is smaller.

[0023]

[0024] Gradient filtering can filter out some water ripple noise in the near area, reduce the complexity of the subsequent filtering process, and at the same time does not affect the edge information of small targets in the distance, retaining the edge features of small targets.

[0025] (2) Noise filtering based on a support vector machine: Some small water ripple noises can be filtered out through the gradient blur algorithm, but water ripple noises with rich features such as waves cannot be effectively filtered out. In this embodiment, through statistical analysis of the area ratios and distances of a large number of ship targets and water ripple noises, as Figure 2As shown, there is an obvious demarcation area in the spatial relationship between the ship and the ripple noise. In this application, an area size classifier based on support vector machine is designed, and through iterative learning, the classification boundary line of the area sizes of the ship target and the noise target is obtained. As shown in Equation (2), x represents the distance between the border of its moving target and the camera, and y represents the proportion of the area of the moving target border to the picture area. According to this classifier, it can be identified which borders belong to the real ship target and which borders are ripple noises, and the ripple noise target boxes are filtered out, further realizing the effect of removing ripple noises.

[0026]

[0027] Among them, the relationship diagram of the proportion of the ship and ripple noise target areas to the distance is shown in Figure 3 as shown.

[0028] The effect diagrams of ripple noise reduction by means of gradient blur filtering and the ripple noise classifier of support vector machine are shown in Figure 4 as shown.

[0029] By Figure 4 it can be obtained that: by combining multiple noise reduction algorithms, in the process of filtering ripple noises from the output results of moving target recognition, the field environment test results show that all water environment noises can be filtered out, and at the same time, it still has the ability to accurately capture ship targets, improving the accuracy of the moving detection algorithm for ship recognition, meeting the application requirements of ship target recognition, and providing an accurate basis for ship monitoring and dispatching.

[0030] Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A ship target detection method based on water environment noise reduction, characterized in that: The method combines two filtering methods: a water pattern noise classifier with gradual fuzzy filtering and a support vector machine; The scheme based on gradual fuzzy filtering is as follows: According to the imaging characteristics of high-depth waterway monitoring, the distribution of water environment noise is mostly in the near area of the monitoring perspective, and at the same distance, there are significant differences in the area between the water environment noise and the ship target area; The size and feature richness of ship targets and water pattern noise in the waterway monitoring image are correlated with the position of the target in the picture. Ship targets with a large area and a lot of water pattern noise are close and located below the monitoring image, while ship targets with a small area and little water pattern noise are far away and located in the upper half of the monitoring image. Therefore, the gradual filtering algorithm is as shown in (1): Ksize = log2h(1), where h represents the ordinate of the pixel point in the image, that is, the filtering template for the water area closer to the bottom of the image is large, and the filtering template for the water area located above the image in the distance is small; Noise filtering based on support vector machine: Fine water ripple noise is filtered out through the gradient blur algorithm, but water ripple noise with rich features such as waves cannot be effectively filtered out. Therefore, statistical analysis is carried out on the variation relationship of the ratio of the area of multiple ship targets to the image area and the ratio of the area of water environment noise to the image area with distance. There is an obvious boundary area in the spatial relationship between ships and water ripple noise. Therefore, an area size classifier based on support vector machine is used, and through iterative learning, the classification boundary line of the area size of ship targets and noise targets is obtained, as shown in Equation (2). , where x represents the distance between the bounding box of the moving target and the camera, and y represents the ratio of the area of the bounding box of the moving target to the screen area. Based on this classifier, it is identified which belong to real ship targets and which belong to water ripple noise, and the bounding box of the water ripple noise target is filtered out to achieve the effect of removing water ripple noise.

Images