Surface Navigation Target Detection and Recognition based on SSD

In order to enhance the perception ability of unmanned ship's surface environment and solve the problem that the target can not be effectively classified and recognized in radar or sensor-based target detection method, this paper collects and manufactures the data set of surface target detection independently, uses machine learning algorithm to detect and recognize multi-size and multi-type targets in ship image, the default target box parameters are initialized by clustering algorithm, which improves the accuracy of the training model in small target detection rate and target box location. The experimental results show that the mAP of the improved model has been upgraded, which can more accurately accomplish the task of multi-target, multi-type and multi-scale surface object detection and recognition, and provide more abundant and effective surface environment information for the automatic driving of unmanned vehicle.