基于MS-YOLOv3的遥感图像小目标检测算法  

Small target detection algorithm in remote sensingimage based on MS-YOLOv3

作　　者：王欣 江涛[1] 何维娟 WANG Xin;JIANG Tao;HE Wei-juan(College of Mathematics and Computer Science,Yunnan Minzu University,Kunming 650500,China)

机构地区：[1]云南民族大学数学与计算机科学学院,云南昆明650500

出　　处：《云南民族大学学报（自然科学版）》2023年第6期764-770,共7页Journal of Yunnan Minzu University:Natural Sciences Edition

基　　金：国家自然科学基金(61363022);云南民族大学研究生创新基金(SJXY-2021-018)。

摘　　要：针对YOLOv3目标检测算法在遥感图像小目标检测中,存在检测准确度低和误检率高的问题,提出了一种基于改进YOLOv3的小目标检测算法:MS-YOLOv3.为了提高小目标特征提取的能力,在原YOLOv3的3个检测尺度基础上,增加了104×104的尺度,有效地掌握了样本的特点;同时在特征提取阶段,把通道注意力模块(MS-CAM)加入原网络中的残差块内,MS-CAM通过两个不同尺度的分支来获得通道注意力,使得语义和尺度不一致的特征能更好地融合,从而更准确地检测出小目标.实验表结果明,MS-YOLOv3算法能够有效抑制复杂背景的干扰,目标区域位置更加准确,检测的精确度更高、误检率更低,更适合遥感图像小目标的检测.Aiming at the low detection accuracy and high false detection rate of YOLOv3 object detection algorithm in remote sensing image small object detection,a small object detection algorithm based on improved YOLOv3 is proposed:MS-YOLOv3.In order to improve the ability of small object feature extraction,based on the three scales of the original YOLOv3,the scale of 104×104 is added to effectively grasp the characteristics of the sample;at the same time,in the feature extraction stage,the channel attention module(MS-CAM)is added to the residual block of the original network.MS-CAM obtains channel attention through two branches of different scales,so that the features with inconsistent semantics and scales can be better fused,so as to detect small objects more accurately.The experimental results show that the MS-YOLOv3 algorithm can effectively suppress the interference of complex backgrounds;the location of the object area is more accurate;the detection accuracy is higher;the false detection rate is lower,and it is more suitable for the detection of small objects in remote sensing images.

关 键 词：遥感图像 YOLOv3 小目标检测 MS-CAM 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]