基于深度学习的水面无人船前方船只图像识别方法  

Image Recognition Method of Ships in front of Unmanned Surface Vessel Based on Deep Learning

作　　者：王贵槐[1] 谢朔 初秀民[3] 洛天骄 WANG Guihuai;XIE Shuo;CHU Xiumin;LUO Tianjiao(Wuhan Technical College of Commtmications, Wuhan 430065, China;Wuhan University of Technology, a. Engineering Research Center for Transportation Safe, b. School of Energy and Power Engineering, c. School of Logistics Engineering, Wuhan 430063, China)

机构地区：[1]武汉交通职业学院,武汉430065 [2]武汉理工大学国家水运安全工程技术研究中心,武汉430063 [3]武汉理工大学能源与动力工程学院,武汉430063 [4]武汉理工大学物流工程学院,武汉430063

出　　处：《船舶工程》2018年第4期19-22,99,共5页Ship Engineering

基　　金：武汉市科技计划项目(2017010201010132)

摘　　要：建立基于图像识别系统的水面无人船(USV)感知平台,采集内河船舶图片数据库建立船只单次多重检测(SSD)深度学习框架,通过使用预训练模型参数调优并微调分类框架实现较高的内河船舶检测准确度。试验结果表明,不同天气状况下的识别算法的查全率和查准率均能保持在70%以上。The USV perception platform based on the image recognition system is established. By collecting inland ship image database, a SSD deep learning framework is built to detect the ships. With the pre-training model parameters tuning and fine-tuning of the classification framework, a high target recognition accuracy of inland ships is achieved. The experimental results show that the recall rate and precision rate of the proposed recognition algorithm can both reach more than 70% under different weather conditions.

关 键 词：深度学习 水面无人船 图像识别 单次多重检测器 

分 类 号：U674.2[交通运输工程—船舶及航道工程] TP391.41[交通运输工程—船舶与海洋工程]