摘要
为解决遥感图像难检测、识别精度低的问题,提出一种基于改进YOLOv5s的遥感目标检测算法。在特征提取网络嵌入Swin transformer模块,实现目标与场景的关系建模,减少误检现象;设计了由三条空洞卷积构造的增大感受野模块,扩大特征图感受野;以跳跃连接的方式引入多尺度特征融合,增强算法对目标尺度变化较大的适应能力,提高融合效率;将Neck部分中的部分标准卷积替换为可变形卷积,强化对目标自身区域和边缘特征的提取能力。在DIOR数据集上,通过消融实验,证明了各改进之处的有效性,所提算法的平均精度均值mAP_(0.5)相较于原模型提升了3.62%,有效提高了遥感目标检测识别精度,证明了改进YOLOv5s算法的有效性,可为解决遥感目标的误检漏检问题提供依据。
In order to solve the problem of difficult detection and low recognition accuracy of remote sensing images,a remote sensing target detection algorithm based on improved YOLOv5s is proposed.Swin transformer module is embedded in the feature extraction network to realize the relationship modeling between the target and the scene and reduce the phenomenon of false detection.An enlarged receptive field module is designed to enlarge the receptive field of the feature map.The multi-scale feature fusion is introduced in the manner of jump connection to enhance the adaptability of the algorithm to the large-scale change of the target scale and improve the fusion efficiency.Some standand convolution in Neck part is replaced with deformable convolution to enhance the ability of extracting the target's own region and edge features.On DIOR data set,the effectiveness of each improvement is proved by ablation experiment.The average accuracy of mAP_(0.5) of the proposed algorithm is 3.62%higher than that of the original model,effectively improving the detection and recognition accuracy of remote sensing targets,which proves the effectiveness of the improved YOLOv5s algorithm.It can provide a basis for solving the problem of false detection and missing detection of remote sensing targets.
作者
陈福明
陈西曲
CHEN Fuming;CHEN Xiqu(School of Electrical and Electronic Engineering,Wuhan Polytechnic University,Wuhan 430023,China)
出处
《武汉轻工大学学报》
CAS
2024年第1期77-85,共9页
Journal of Wuhan Polytechnic University
