基于级联检测与超分辨率的坝体杂草株高测量算法

打开文本图片集

中图分类号：TP391.41 文献标识码：A 文章编号：1006-8228（2026）01-27-06

Abstract：Toaddresschalengesindamweedheightmeasurement—suchassmalltargetsize，blurrdreferencescales，and inaccuratelocalization—thispaperproposesameasurementalgorithmbasedoncascadeddetectionandsuper-resolution.Themethod adoptsa"detection—super-resolution—redetection"cascadedframework：inthefirststage，animprovedYOLOv8-Rnetworkis employedtolocatetheentirereferencerulerwithinscenescontainingweds；inthesecondstage，thedetailsofthecropedruler regionareenhancedusingtheReal-ESRGANsupe-resolutionnetwork，andthentheregionisfedintoaYOLOv8-Bnetworkfor preciserecognitionof individualcolorblocksontheruler.ExperimentalresultsshowthattheproposedalgorithmachievesmAP50 scoresof 99.2% and 96.9% onthe two independent detection tasks：ruler （Ruler） and color block （Block），respectively. Furthermore，among validsamples，the meanabsoluteeror between thecomputedweed heightsandmanualreadingsisonly 0.65cm This approach enables high-precision，low-cost，automated monitoring of weed height on dam surfaces.

keywords：Cascaded Detection； Super-Resolution;Deep Learning;Weed HeightMeasurement

0引言

水库作为水资源调控与防洪体系中的关键基础设施，其坝体安全直接关系到工程效益与周边区域的安全保障。（剩余7761字）