基于改进YOLO和迁移学习的水下鱼类目标实时检测.pdf
需积分: 26 185 浏览量
2020-03-31
19:21:49
上传
评论 5
收藏 1.4MB PDF 举报
摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇 摇
基
基
于改进 YOLO 和迁移学习的
水下鱼类目标实时检测
李庆忠
1
摇 摇 李宜兵
1
摇 摇 牛摇 炯
1
摘摇 要摇 为了实现非限制环境中水下机器人基于视频图像的水下鱼类目标快速检测,提出基于改进 YOLO 和迁移
学习的水下鱼类目标实时检测算法. 针对 YOLO 网络的不足,设计适合水下机器人嵌入式系统计算能力的精简
YOLO 网络(Underwater鄄YOLO). 利用迁移学习方法训练 Underwater鄄YOLO 网络,克服海底鱼类已知样本集有限的
限制. 利用基于限制对比度自适应直方图均衡化的水下图像增强预处理算法,克服水下图像的降质问题. 利用基于
帧间图像结构相似度的选择性网络前向计算策略,提高视频帧检测速率. 实验表明,文中算法能实现在非限制环境
下海底鱼类目标的实时检测. 相比 YOLO,文中算法对海底鱼类小目标和重叠目标具有更好的检测性能.
关键词摇 卷积神经网络, 深度学习, 鱼类目标检测, 单级式目标检测算法( YOLO), 迁移学习
引用格式摇 李庆忠,李宜兵,牛 炯. 基于改进 YOLO 和迁移学习的水下鱼类目标实时检测. 模式识别与人工智能,
2019, 32(3): 193-203.
DOI摇 10. 16451 / j. cnki. issn1003鄄6059. 201903001摇 摇 摇 摇 摇 摇 中图法分类号摇 TP 181
Real鄄Time Detection of Underwater Fish Based on
Improved YOLO and Transfer Learning
LI Qingzhong
1
, LI Yibing
1
, NIU Jiong
1
ABSTRACT 摇 To fast detect underwater fish in unrestricted underwater environment based on
underwater video collected by underwater robots, a real鄄time detection algorithm for underwater fish based
on improved you only look once(YOLO) and transfer learning is proposed. Firstly, an underwater鄄YOLO
for the embedding computer system of underwater robots is designed to overcome the shortcomings of
traditional YOLO. Then, transfer learning strategy is employed to train the underwater鄄YOLO network
and alleviate the limitation of known underwater fish samples. A preprocessing algorithm based on
contrast limited adaptive histogram equalization is proposed to overcome the problem of underwater image
degradation. Finally, a video frame selection method for foreground computation of underwater鄄YOLO
based on structure similarity between inter鄄frames is proposed to increase the detection frame rate. The
experimental results show that the proposed algorithm achieves the goal of real鄄time detection of
underwater fish in unconstrained underwater environment. Compared with the traditional YOLO, the
proposed underwater鄄YOLO generates better detection performance in complex scenes with small fish and
overlapped fishes.
Key Words摇 Convolutional Neural Network, Deep Learning, Fish Target Detection, You Only Look
Once( YOLO), Transfer Learning
收稿日期:2018-11-14;录用日期:2019-03-13
Manuscript received November 14, 2018;
accepted March 13, 2019
国家重点研发计划项目(No. 2017YFC1405202)、国家自然科
学基金项目(No. 61132005)、海洋公益性行业科研专项(No.
201605002)
Supported by National Key R&D Plan of China(No. 2017YFC14
05202), National Natural Science Foundation of China ( No.
61132005 ), National Marine Technology Program for Public
Welfare of China (No. 201605002)
本文责任编委 叶东毅
Recommended by Associate Editor YE Dongyi
1. 中国海洋大学 工程学院摇 青岛 266100
1. College of Engineering, Ocean University of China, Qingdao
266100
第 32 卷摇 第 3 期 模式识别与人工智能 Vol. 32摇 No. 3
2019 年 3 月 Pattern Recognition and Artificial Intelligence Mar. 摇 2019
摇 摇 摇 摇 摇
摇 摇 摇 摇 摇
摇 摇 摇 摇
摇 摇 摇 摇 摇
摇 摇 摇 摇 摇
摇 摇 摇 摇 摇
摇 摇 摇 摇 摇
摇
《模式识别与人工智能》电子版
剩余10页未读，继续阅读
评论0
最新资源