作者：郭海艳1, 程亮2,3, 杨春利3, 刘斌3, 王磊刚3, 闫雪梅3, 熊锐1, 陈朋鹏1, 何赟泽1
作者单位：1. 湖南大学电气与信息工程学院, 湖南 长沙 410006;
2. 江苏海洋大学海洋工程学院, 江苏 连云港 222005;
3. 珠海云洲智能科技有限公司, 广东 珠海 519085
关键词：水面无人艇;水上目标和船舶检测;yolov5l;网络级联;网络部署;aspp-pool
摘要：
搭载视觉智能感知系统的水面无人艇能够对水上目标有效识别，其在水面勘探、自主搜救等领域有着广泛的应用；对水上目标和船舶的检测方法进行研究，其主要工作包括：第一，构建水面无人艇的数据采集和边缘计算平台；第二，构建水上目标数据集，总计图片3437张，28个类别，其中8个类别为船舶分类；第三，分别训练水上目标检测和船舶检测的yolov5l网络并进行测试，并将网络部署到边缘计算平台，水上目标检测网络map为78.23%，船只分类网络map为85.16%；第四，基于检测框匹配的方式实现水上目标检测网络和船只分类网络的级联工作，级联网络map达到78.58%；第五，引入aspp-pool模块，并训练改进后的网络，水上目标检测网络map提高1.06%，级联网络map提高0.76%。
research on target detection and ship classification system for unmanned surface vessels
guo haiyan1, cheng liang2,3, yang chunli3, liu bin3, wang leigang3, yan xuemei3, xiong rui1, chen pengpeng1, he yunze1
1. college of electrical and information engineering, hunan university, changsha 410006, china;
2. school of ocean engineering, jiangsu ocean university, lianyungang 222005, china;
3. zhuhai yunzhou intelligent technology co., ltd., zhuhai 519085, china
abstract: the unmanned surface vessels equipped with visual intelligent sensing system can effectively identify water targets, which is widely used in the fields of surface exploration, autonomous search and rescue and so on; the detection methods of water targets and ships are studied; the main work includes the following points: firstly, the data acquisition and edge calculation platform of unmanned surface vessels is constructed; secondly, build a water target data set, with a total of 3437 pictures and 28 categories, of which 8 categories are ship classification; thirdly, the yolov5l network for water target detection and ship detection is trained and tested respectively, and the network is deployed to the edge computing platform; the water target detection network map is 78.23%, and the ship classification network map is 85.16%; fourthly, the cascade work of water target detection network and ship classification network is realized based on detection frame matching, and the cascade network map reaches 78.58%; fifthly, the aspp-pool module is introduced and the improved network training network is used; the water target detection network map is improved by 1.06% and the cascade network map is improved by 0.76%;
keywords: unmanned surface vessels;detection of water targets and ships;yolov5l;network cascaded;network deployment;aspp-pool
2023, 49(6):114-121 收稿日期: 2022-01-18;收到修改稿日期: 2022-03-18
基金项目: 湖南省自然科学基金重大项目（s2021jjzdxm0022）；湖南省重点研发计划（s2021gczdyf0800）；珠海云洲智能科技有限公司委托课题（h202191400326）
作者简介: 郭海艳(2000-),女,湖南衡阳市人,硕士研究生,专业方向为图像处理、深度学习
参考文献
[1] 梁秀梅. 无人艇视觉系统目标图像特征提取与识别技术研究[d]. 哈尔滨: 哈尔滨工程大学, 2013.
[2] 熊勇, 余嘉俊, 张加, 等. 无人艇研究进展及发展方向[j]. 船舶工程, 2020, 42(2): 12-9
[3] 郭雨青, 曾庆军, 夏楠, 等. 图像增强水下自主机器人目标识别研究[j]. 中国测试, 2021, 47(11): 47-52
[4] 高继东, 焦鑫, 刘全周, 等. 机器视觉与毫米波雷达信息融合的车辆检测技术[j]. 中国测试, 2021, 47(10): 33-40
[5] 彭伟康, 陈爱军, 吴东明, 等. 基于改进faster r-cnn的水准泡缺陷检测方法[j]. 中国测试, 2021, 47(7): 6-12
[6] 刘梦佳. 基于cnn的搜救无人艇水面多目标检测实验[j]. 武汉理工大学学报:交通科学与工程版, 2019, 43(5): 5
[7] zeng l, sun b, zhu d. underwater target detection based on faster r-cnn and adversarial occlusion network[j]. engineering applications of artificial intelligence, 2021, 100(4): 104190
[8] girshick r, donahue j, darrell t, et al. rich feature hierarchies for accurate object detection and semantic segmentation[c]// proceedings of 2014 ieee conference on computer vision and pattern recognition (cvpr), 2014: 580-587.
[9] redmon j, divvala s, girshick r, et al. you only look once: unified, real-time object detection[c]//proceedings of the ieee conference on computer vision & pattern recognition. 2016: 779-788.
[10] redmon j, farhadi a. yolo9000: better, faster, stronger[c]//proceedings of the ieee conference on computer vision & pattern recognition, 2017: 7263-7271.
[11] redmon j, farhadi a. yolov3: an incremental improvement[eb/ol]. [2018-04-08]. https://arxiv.org/pdf/1804.02767.pdf.
[12] bochkovskiy a, wang c y, liao h. yolov4: optimal speed and accuracy of object detection[eb/ol]. [2020-04-23]. https://arxiv.org/pdf/2004.10934.pdf.
[13] 高照. 基于深度学习的遥感图像目标检测方法研究 [d]. 青岛: 山东科技大学, 2020.
[14] chen lc, papandreou g, kokkinos i, et al. deeplab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs[j]. computer vision and pattern recognition, 2017, 40(4): 834-48
[15] chan t f, vese l a j i t o i p. active contours without edges[j]. ieee transactions on image processing : a publication of the ieee signal processing society, 2001, 10(2): 266-77
[16] lin t y, goyal p, girshick r, et al. focal loss for dense object detection[j]. ieee transactions on pattern analysis & machine intelligence, 2017(99): 2999-3007
