Research On Detection And Recognition Algorithm For The Surface Object

The total economy of marine grows rapidly in recent years and becomes an important part of the national economic development. Marine equipment manufacturing and harbor facilities play an important role in the development of marine economy. As an intelligent marine equipment, the USV(Unmanned Surface Vehicles) attracts the attention all over the world and develops rapidly. In this paper, we focus on the issue of surface object detection and recognition which contains two applications in practice. The first application is the research on surface object detection and recognition based on the USV. The second application is the research on harbor ship detection in the infrared image.For the research on surface object detection and recognition based on the USV, as the climate is quite complex, the background is complicated and the object category is various, it will be very difficult if we detect the object on the original image directly, furthermore, speed and accuracy can't be guaranteed. In this paper, we propose two different methods to solve this problem. The first one is that combines the objectness and saliency. Firstly, we use the objectness to get the object proposals. The proposals may have many false alarms at this time. Secondly, we apply the idea of saliency to calculate the salient map. Finally, fuse the result of objectness and saliency to remove the object proposals with false alarms. This algorithm is quite general without any specific object type. Another approach is using deep learning to solve this problem. Deep learning can give the definite category and confidence at the same time. Compared to other existing object detection and recognition algorithms, the methods proposed in this pater have improved the accuracy and speed which make great significance for the USV to avoid the obstacle automatically and navigate autonomously.For the research on harbor ship detection in the infrared image, because of the characteristics of infrared image, it is very difficult to use the edge and texture information of the ship for detection. The background in the harbor is quite complex containing ship, seawater, buildings, land and so on. What is more, the pose, location, size, number of the ship is uncertain. Taking above difficulties into consideration, the method proposed in this paper can be divided into two basic steps: harbor region matching and ship detection. Firstly, find the harbor region by template matching. Secondly, detect the ship in the harbor region. By this way, we reduce the search space and decrease the interference of the complicated background in order to improve the efficiency and accuracy of the algorithm. Finally, we have done a lot of experiments to show that our approach has a high accuracy rate and low false alarm rate.