复杂场景下深度表示的SAR船舶目标检测算法  

SAR Target Detection Algorithm for Depth Representation in Complex Scenes

作　　者：袁国文 黄鸿勇[1] YUAN Guowen;HUANG Hongyong(School of Intelligent Manufacturing,Guangdong Polytechnic,Foshan,Guangdong 528000,China)

机构地区：[1]广东职业技术学院智能制造学院,广东佛山528000

出　　处：《自动化应用》2023年第23期161-163,共3页Automation Application

基　　金：广东职业技术学院校级常规科研项目(XJKY202228)。

摘　　要：为改善合成孔径雷达(SAR)船舶在复杂背景下目标检测精度低、速度慢的问题,本文在YOLOv5系列算法的基础上,提出一种复杂场景下深度表示的SAR船舶目标检测算法。该算法通过改进YOLOv5中Backbone层的整体结构,修改Conv层与BottleneckCSP层的激活函数,提升非线性问题的处理能力。结果表明,改进后的YOLOv5算法精度可达93.8%,优于YOLO系列的其他算法,提高了SAR船舶目标检测的精度与速度。In order to improve the low accuracy and slow speed of target detection of synthetic aperture radar(SAR)ships in complex background,this paper proposes a SAR ship target detection algorithm based on depth representation in complex scenes on the basis of YOLOv5 series of algorithms.The algorithm improves the overall structure of the backbone layer in YOLOv5,modifies the Activation function of the Conv layer and Bottleneck CSP layer,and improves the processing ability of nonlinear problems.The results show that the improved YOLOv5 algorithm has an accuracy of 93.8%,which is superior to other YOLO series algorithms.It improves the accuracy and speed of SAR ship target detection.

关 键 词：SAR图像 目标检测 YOLOv5 激活函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]