植被指数引导的高光谱影像小麦种植区提取方法研究  

Research on Wheat Planting Areas Extraction Method from Hyperspectral Images Guided by Vegetation Index

作　　者：田婷婷 杜斌 谭美淋 岳云峰 黄仡凡 张帅 TIAN Tingting;DU Bin;TAN Meilin;YUE Yunfeng;HUANG Yifan;ZHANG Shuai(Inner Mongolia Autonomous Region Surveying and Mapping Geographic Information Center,Hohhot 010000,China;Fujian Jingwei Digital Technology Co.,Ltd.,Fuzhou 350000,China;StarvisionWorld Technology Corporation,Wuhan 430000,China)

机构地区：[1]内蒙古自治区测绘地理信息中心,内蒙古呼和浩特010000 [2]福建省经纬数字科技有限公司,福建福州350000 [3]武汉星视寰宇科技有限公司,湖北武汉430000

出　　处：《地理空间信息》2025年第4期59-62,共4页Geospatial Information

基　　金：内蒙古自治区测绘地理信息中心自选项目(NMGZCS-C-F-220342)。

摘　　要：农作物对于不同波段的反射和吸收呈现出显著的光谱特征,这使得高光谱图像在农作物分类中具备天然的优势。为了解决高光谱遥感影像农作物提取中的空间信息缺失和光谱响应不一致问题。提出多种植被指数耦合的农作物提取网络,利用植被指数构建光谱特征库,辅助网络进行农作物提取,通过引入尺度间的注意力机制和交叉通道增强模块,弥补空间描述信息和通道相关性信息的不足。以内蒙古典型作物小麦为实验对象进行研究,结果显示该方法在农作物提取方面表现出显著性能.Crops exhibit significant spectral characteristics in different bands,providing a natural advantage for crop classification in hyperspectral images.To address spatial information gaps and inconsistent spectral responses in crop extraction from hyperspectral remote sensing images,we proposed a crop extraction network that coupled multiple vegetation indices.We utilized vegetation indices to construct a spectral feature library,assisting in crop extraction.By introducing a scale-aware attention mechanism and a cross-channel enhancement module,we addressed deficiencies in spatial description and channel correlation information.Taking wheat,a typical crop in Inner Mongolia,as the experimental subject,the result demonstrates the significant performance of this method in crop extraction.

关 键 词：植被指数 高光谱影像 农作物提取 深度学习 

分 类 号：P237[天文地球—摄影测量与遥感]