融合视觉中心机制和并行补丁感知的遥感图像检测算法

Remote sensing image detection algorithm integrating visual center mechanism and parallel patch perception

针对遥感图像存在复杂背景干扰、目标多尺度差异和微小目标提取难的问题,本文基于YOLOv7-tiny模型提出一种融合视觉中心机制和并行补丁感知的遥感图像检测算法。该算法一是引入显式视觉中心机制,构建像素点间的长距离依赖关系,丰富图像的整体语义信息,同时提升对目标纹理的提取性能;二是改进并行补丁感知模块,调整特征提取感受野,以适应不同目标尺度;三是设计多尺度特征融合模块,实现对多层特征的高效融合,提升模型推理速度。在公共数据集RSOD上进行实验,所提算法的准确率、召回率和平均准确率均值相较YOLOv7-tiny分别提升1.5%、2.4%和2.4%,此外在NWPUVHR-10和DOTA数据集上进行泛化性验证,结果表明本文算法具备较强的泛化性能。通过与不同算法对比分析,进一步体现本文算法性能的优越性。

To address the challenges of complex background interference,multi-scale differences in targets,and the difficulty in extracting small targets from remote sensing images,this paper proposes a remote sensing image detection algorithm based on the YOLOv7-tiny model that integrates the visual center mechanism and parallel patch perception.Firstly,the algorithm introduces an explicit visual center mechanism to establish long-distance dependencies between pixels,enriching the overall semantic information of the image and improving the extraction performance of target textures.Secondly,it improves the parallel patch perception module by adjusting the feature extraction receptive fields to adapt to different target scales.Thirdly,a multi-scale feature fusion module is designed to efficiently fuse multi-layer features,thereby improving the model's inference speed.Experimental results on the RSOD dataset show that the proposed algorithm achieves improvements over YOLOv7-tiny in terms of precision,recall,and mean average precision by 1.5%,2.4%,and 2.4%,respectively.Additionally,validation on the NWPU VHR-10 and DOTA datasets confirms the strong generalization performance of the proposed algorithm.Comparative analysis with other algorithms further demonstrates the superior performance of the proposed approach.