基于Segformer网络的地块尺度作物种植结构精细化识别与分类  

Refined identification and classification of crop planting structure at plot scale based on Segformer network

作　　者：顾余庆 李晓文 曹伟 王亚华[2,3] 周鑫鑫[3,4] 赵碧 GU Yu-qing;LI Xiao-wen;CAO Wei;WANG Ya-hua;ZHOU Xin-xin;ZHAO Bi(Nanjing Guotu Information Industry Co.,Ltd.,Nanjing 210036,China;School of Geography,Nanjing Normal University,Nanjing 210023,China;Key Laboratory of Virtual Geographic Environment,Nanjing Normal University,Ministry of Education,Nanjing 210023,China;School of Geographic and Biologic Information,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京国图信息产业有限公司,江苏南京210036 [2]南京师范大学地理科学学院,江苏南京210023 [3]南京师范大学虚拟地理环境教育部重点实验室,江苏南京210023 [4]南京邮电大学地理与生物信息学院,江苏南京210023

出　　处：《江苏农业学报》2024年第2期293-302,共10页Jiangsu Journal of Agricultural Sciences

基　　金：国家自然科学基金项目(42201504、41971404);江苏省自然资源厅科技项目(2021013)。

摘　　要：防止耕地“非粮化”、稳定粮食生产是中国粮食安全的基石。为实现地块破碎化地区作物类型及种植结构精细化识别和分类,本研究以江苏省泰兴市为研究区,基于高分辨率遥感影像和多尺度融合特征显著的Segformer语义分割模型,实现地块尺度的耕地信息精细化提取;同时结合多源遥感数据构建主要植被类型归一化植被指数(NDVI)时序曲线及植被生长关键时间节点的光谱反射特征,开展地块尺度的作物种植结构分类。结果表明:基于Segformer模型的分割方法可有效识别耕地,F1系数达92.4%;基于主要植被类型多时相NDVI时序特征及植被生长关键时间节点光谱反射特征的作物种植结构分类方法能够实现地块尺度的种植结构分类,总体分类精度达82.38%。因此,本研究建立的方法可有效实现地块尺度耕地信息的精细化提取及种植结构识别和分类,为耕地保护提供技术支持。Preventing the“non-grain”of cultivated land and stabilizing food production are the cornerstones of China’s food security.In order to realize the fine identification and classification of crop types and planting structure in the area of land fragmentation,this study took Taixing City,Jiangsu province as the research area,and realized the fine extraction of cultivated land information at the plot scale based on the high-resolution remote sensing images and the Segformer semantic segmentation model with significant multi-spatial scale fusion features.At the same time,the normalized difference vegetation index(NDVI)time series curve of the main vegetation types and the spectral reflectance characteristics at the key time nodes of vegetation growth were constructed by combining multi-source remote sensing data,and the classification of crop planting structure at the plot scale was carried out.The results showed that the segmentation method based on Segformer model could effectively identify cultivated land,and the F1 was 92.4%.The classification method of crop planting structure based on multi-temporal NDVI time series characteristics of main vegetation types and spectral reflection characteristics at key time nodes of vegetation growth could realize the classification of planting structure at plot scale,and the overall classification accuracy was 82.38%.Therefore,the method established in this study could effectively realize the fine extraction of cultivated land information at the plot scale and the identification and classification of planting structure,and provide technical support for cultivated land protection.

关 键 词：种植制度 地块尺度 精细化识别和分类 遥感 

分 类 号：S127[农业科学—农业基础科学]