基于深度学习的盐碱地遥感影像反演模型构建

打开文本图片集

关键词：深度学习；盐碱地；遥感影像；反演模型；注意力机制中图分类号：TP753 文献标识码：A文章编号：0439-8114（2025）06-0185-05DOI：10.14088/j.cnki.issn0439-8114.2025.06.031开放科学（资源服务）标识码（OSID）：

Construction of a remote sensing image inversion model for saline-alkali land based on deep learning

BANRuo-nan，DONGLuan

（CollgeofComputerand Information Engineering，Xinjiang Agricultural University，Urumqi 83oo52，China

Abstract：Toaddresstheinversionprobemofremotesensingimagesofsaline-alkalilandintheTailanRiverIrgationDistrictof Wensu CuntyAksurefecture，Xinjiang，animageclasificationmodelbasedontheointRendnetworkintegratedwiththeCAttentionmechanismwas proposed.The modelusedDeepLabV3asthebackbone networkandincorporatedtheCoTAtentionmodule to enhancethefeatureextractioncapabityoftenetwork.Toverifytheclasificationperformanceoftheimprovedntwork，experiments wereconductedusing Sentinel-2remotesensing imagesfromtheTailanRiverIrigationDistrict.TheresultsshowedthatthePointRend model achieved good performance metrics，with pixel accuracy，mean intersection over union （MIoU），and FI reaching 89 % ， 88% ，and 88% ，respectively.Compared withthe original PointRend model without the CoTAtention mechanism，the improved PointRend model demonstrated enhanced metrics，with pixel accuracy， MIoU ，and FI increasing by 3， 2 ，and 3 percentage points，respectively.TheimprovedPointRendmodelsignficantlyenhancedtheabiltytocapturefinedetailsandrefineedgesinsegmentation tasks，thereby improving overall classification accuracy.

Key Words：deep learning；saline-alkaliland;remote sensing images；inversion model;attention mechanism

盐碱地是中国重要的后备耕地资源。（剩余7522字）