基于多尺度特征融合的地理测绘影像目标检测  

Target Detection of Geographic Mapping Image Based on Multi-scale Feature Fusion

作　　者：李睿 李亚洲 赵建文 周卫波 LI Rui;LI Yazhou;ZHAO Jianwen;ZHOU Weibo(State Grid Shandong Electric Power Construction Company,Jinan 250000,China)

机构地区：[1]国网山东省电力公司建设公司,山东济南250000

出　　处：《林业调查规划》2024年第4期188-194,共7页Forest Inventory and Planning

基　　金：山东省电力公司科技项目(520632220002).

摘　　要：为了提高对地理测绘目标的检测准确度,设计了基于多尺度特征融合的地理测绘影像目标检测方法。初步提取地理测绘遥感影像的边缘信息,并计算其边缘密度与边缘分布情况,通过增强边缘信息实现对遥感影像的预处理,得到更明确的影像边缘信息;利用梯度采样法建立下降金字塔影像,并融合多尺度特征,为后续的目标提取提供更准确、特征更明显的信息;根据特征融合结果,采用深度卷积网络实现对地理测绘影像目标的有效检测。结果表明,应用该方法,检测结果的准确率、召回率和F 1分数数值均较高,检测耗时也维持在较低的数值范围,该方法可明显提高目标检测效果。In order to improve the detection accuracy of geographic mapping targets,a target detection method of geographic mapping images based on multi-scale feature fusion was designed.The edge information of the remote sensing image of geographical mapping was preliminarily extracted,and its edge density and edge distribution were calculated.Through enhancing the edge information,the remote sensing image was preprocessed to obtain more clear image edge information.Gradient sampling method was used to establish the descending pyramid image to provide more accurate and distinct information for subsequent target extraction by integrating multi-scale feature.According to the feature fusion results,the deep convolution network was used to effectively detect the geographic mapping image objects.The experimental results showed that the accuracy,recall and F 1 score of the detection results were high after the application of this method,and the detection time was also maintained in a lower numerical range,indicating that the method significantly improved the detection effect for targets.

关 键 词：目标检测 地理测绘影像 边缘信息 多尺度特征 深度卷积网络 检测耗时 

分 类 号：S237[农业科学—农业机械化工程] TP753[农业科学—农业工程]