基于改进YOLOv7的光学遥感图像船舶旋转目标检测  

Optical Remote Sensing Image Ship Rotating Object Detection Based on Improved YOLOv7

作　　者：焦仕昂 罗亮 杨萌[3] 翟宏睿 刘维勤 JIAO Shiang;LUO Liang;YANG Meng;ZHAI Hongrui;LIU Weiqin(Key Laboratory of High Performance Ship Technology,Ministry of Education,Wuhan 430063,China;School of Naval Architecture Ocean and Energy Power Engineering,Wuhan University of Technology,Wuhan 430063,China;China Ship Research and Design Center,Wuhan 430064,China)

机构地区：[1]高性能船舶技术教育部重点实验室,武汉430063 [2]武汉理工大学船海与能源动力工程学院,武汉430063 [3]中国舰船研究设计中心,武汉430064

出　　处：《武汉理工大学学报（交通科学与工程版）》2024年第5期903-908,共6页Journal of Wuhan University of Technology（Transportation Science & Engineering）

基　　金：国防科工局国防基础科研计划项目(JCKY2020206B037)。

摘　　要：针对在遥感图像复杂背景下船舶目标尺度变化大以及方向任意的特点,提出一种基于YOLOv7改进的光学遥感图像船舶旋转目标检测算法.在骨干网络中融入全局注意力机制(GAM),使模型在复杂背景中更加关注船舶对象;在颈部加入自适应空间特征融合模块(ASFF),消除了特征金字塔内部特征信息不一致的问题;针对船舶比例细长且方向任意的特点,加入了高斯建模表示的旋转框检测,提高了定位精度并保留了船舶目标方向信息.结果表明:改进的算法在保证检测速度的同时,在HRSC2016数据集取得了90.73%的检测精度,检测效果也有所提升.Aiming at the characteristics of large scale change and arbitrary direction of ship targets in the complex background of remote sensing images,an improved ship rotating target detection algorithm based on YOLOv7 was proposed.By incorporating the Global Attention Mechanism(GAM)into the backbone network,the model was guaranteed to pay more attention to the ship objects in the complex background.By adding an adaptive spatial feature fusion module(ASFF)to the neck,the problem of inconsistent feature information in the feature pyramid was eliminated.Aiming at the characteristics of slender ship scale and arbitrary direction,by adding rotating frame detection represented by Gaussian modeling,the positioning accuracy was improved and the ship target direction information was retained.The results show that the improved algorithm achieves 90.73%detection accuracy in HRSC2016 data set while ensuring the detection speed,and the detection effect is also improved.

关 键 词：YOLOv7 遥感图像检测 船舶旋转目标检测 注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]