基于卷积神经网络的YOLOv3(You Only Look Once v3)目标检测算法设计了一种基于目标检测及模糊匹配机制的非标船牌识别软件系统,并通过在船舶视频监控系统中的实际应用,验证了该船牌识别系统的可行性,提高了船舶铭牌识别的可靠性。YOLOv3目标检测算法将检测简化为一个回归问题,通过仅仅一个网络,就能从图像中得到物体的类别与概率,确保了识别的准确性与实时性。针对船舶非标铭牌锈蚀、遮挡等问题,创新点是在基于YOLOv3的非标船牌识别系统的实现框架之上,设计船名有限中文库与模糊匹配机制,有效解决了船牌识别准确率过低的问题,取得了较好的识别效果。
In this paper,a non standard ship identification characters detector based on object detection and fuzzy matching mechanism was designed according to YOLOv3. Through the application of ship video monitoring system,the ship identification system is verified and improves the reliability of ship identification. YOLOv3 frames object detection simplify detection to a regression problem to obtain category of objects and associated class probabilities via one single network. Therefore,the accuracy and real time of the detection are ensured. For the problem of rust and occlusion of the non standard ship identification characters,the innovation point is designing a limited Chinese library of ship names and fuzzy matching mechanism on the basis of the detection system,thus effectively improving the accuracy of detection.
ZHOU Yi, ZHU Qi-rui, XIE Hai-Cheng, YANG Jian-feng. Non standard ship identification characters detectionbased on target detection and fuzzy matching[J]. LASER & INFRARED,2021,51(11):1526~1530