| Ship detection is widely used in the maritime field and plays a very important role in guaranteeing navigation safety,shipping trade,water management and ship scheduling.Due to the mainstream physical RF signal ship detection methods,it is difficult to detect illegal ships and camouflaged ships in key waters.Relying on the embedded ship face detection project of the research group and Shanghai Dingliang Optoelectronics Technology Co.,Ltd.,this paper researches and implements a computer vision-based ship face detection network and a prototype of a ship detection management system suitable for deployment in edge devices.First,this paper proposes an improved YOLOv4 ship face detection network,which achieves good network detection speed and detection accuracy.Specifically,by replacing the backbone network and using depthwise separable convolution to lighten the network,this paper effectively reduces the amount of network parameters and improves the speed of forward inference.In order to make up for the loss of detection accuracy caused by network lightweight,this paper proposes a lightweight residual-based channel attention module LCARM to enhance network feature extraction capabilities.The experimental results show that compared with YOLOv4,the ship face detection network proposed in this paper reduces the number of parameters by 82.58%,increases the forward inference speed by 53.6%,and achieves an average accuracy of 62.75% m AP@0.75 and 94%m AP@0.5.This network has good real-time and accuracy,and can be deployed on edge devices or servers with limited computing power for ship face detection.In addition,due to the lack of boat face detection datasets that meet the classification criteria.Through data enhancement technology,this paper builds a boat face dataset with 10 categories and20216 samples for model training.The dataset construction method in this paper is applicable to the construction of other general object datasets.Finally,according to the relevant requirements of the port and shipping department,this paper designs and implements a prototype of a ship detection management system based on computer vision.The system takes the ship face detection network as the core to realize efficient and accurate ship detection,digital and visualized ship scheduling functions and port berth monitoring functions,reducing the workload of port operators and reducing the safety risks of navigation in key waters. |
