基于卷积神经网络的高分辨率遥感影像目标边界提取方法

打开文本图片集

关键词：卷积神经网络；高分辨率遥感影像；目标边界提取；深层特征；特征增强；边界概率图中图分类号：TN911.73-34；TP391.41 文献标识码：A 文章编号：1004-373X（2026）01-0049-05

High resolution remote sensing image object boundary extraction method basedonconvolutional neuralnetwork

WANG Xiaohong1，² （1.School of Intelligence Science and Engineering，Qinghai Minzu University，Xining81oooo，China; 2.Geomatics TechnologyandApplicationKeyLaboratoryofQinghaiProvince，Xining81oooo，China）

Abstract：Inviewoftheimpactoffactorssuchasoccusionandrotationonobjectboundaryextractioninhigh-resolution remotesensingimages，aCNN-basedmethodforextractingobjectboundariesfromhigh-resolutionremotesensingimagesis proposed.Thehigh-resolutionremotesensing imageobjectboundaryextractionframework isimplemented byCNN，onthebasis of which，afeatureenhancementmoduleisintroducedtoavoidinsuficientrepresentationofsemanticinformationandlosof detailinformationintheshalowanddeepfeatureextractionofnetworkobjectboundaries.Thenetworklossfunctionis optimized，andhebjectboundaryapispreprocessdandthenconvertedintoaprobabilityapofboundaryinforationnd thenathresholdvalueissettoexcludeuncertainpixels，soastoenhanetherobustnessandaccuracyofmodelobjectboundary extraction.Theexperimentalresultsshowthattheproposed methodcanachieveacurateextractionofobjectboundaries，andis noteasilyaffectedbyremotesensingimagerotation.Inadition，ithasexcelentabilityofobjectboundaryextractionuder different degrees of occlusion.

Keywords：CNN；high resolutionremote sensing image;objectboundary extraction；deep feature;featureenhancement; boundary probability map

0 引言

高分辨率遥感影像作为现代地理信息获取的重要手段，其详细且精确的数据为环境规划、地理分析、资源勘探、灾害评估等多个领域提供了不可或缺的信息支持。（剩余5917字）