摘要
为了提高海上无人艇的舰船目标检测精度和速率,本文基于深度学习方法,利用卷积神经网络、区域建议网络及Fast R-CNN检测框架构建了舰船检测系统。该系统通过共享的卷积神经网络提取特征;通过区域建议网络生成候选区域;通过Fast R-CNN框架实现目标检测识别,从而实现端到端的舰船目标检测。实验结果表明,相比于传统机器学习目标检测算法,该舰船检测系统在检测精度及检测速率上均有大幅提高,达到83.79%的准确率及0.05 s/帧的检测速率。本文的舰船检测系统在检测精度及速率上均表现优异,满足了水面无人艇的工作要求。
In order to improve the detection speed and accuracy of ship target for unmanned surface vehicle, a ship detection system is constructed using convolutional neural networks, region proposal networks and Fast R-CNN detection framework based on deep learning method in this paper. The ship features are extracted by shared convolutional neural networks;the candidate region is generated through region proposal networks;and the target detection and recognition are realized through the Fast R-CNN framework to achieve end to end ship target detection. The experimental results show that,compared with the traditional machine learning algorithms for target detection, the precision and the detection rate of the proposed system has been greatly improved, the mean average precision reached 83.79% and the detection rate reached 0.05seconds/frame. The ship detection system in this paper has excellent performance in detection accuracy and speed, and meets the actual requirements of unmanned surface vehicles.
作者
袁明新
张丽民
朱友帅
姜烽
申燚
YUAN Ming-xin;ZHANG Li-min;ZHU You-shuai;JIANG Feng;SHEN Yi(School of Mechanical Engineering,Jiangsu University of Science and Technology,Zhenjiang 212003,China;Zhangjiagang Industrial Technology Research Institute,Jiangsu University of Science and Technology,Zhangjiagang 215600,China)
出处
《舰船科学技术》
北大核心
2019年第1期111-115,124,共6页
Ship Science and Technology
基金
国家自然科学基金资助项目(61105071)
张家港江苏科技大学产业技术研究院自主产业化资助项目(509914003)
关键词
舰船
目标检测
深度学习
区域建议网络
卷积神经网络
ship
target detection
deep learning
region proposal networks
convolutional neural networks
