多时相遥感影像检测平乐县晚稻种植面积变化  

Detection of late rice's planting area change in Pingle County based on multi-temporal remote sensing images

作　　者：黄维[1,2] 黄进良[1] 王立辉[1] 胡砚霞[1,2] 韩鹏鹏 王久玲[1,2] 

机构地区：[1]中国科学院测量与地球物理研究所,武汉430077 [2]中国科学院大学,北京100049

出　　处：《农业工程学报》2014年第21期174-183,共10页Transactions of the Chinese Society of Agricultural Engineering

基　　金：中科院战略性先导科技专项-应对气候变化的碳收支认证及相关问题(XDA05050107)

摘　　要：为检测中国主要的粮食作物水稻的种植面积变化,该文以广西平乐县为例,利用多时相陆地卫星专题成像仪(landsat thematic mapper)影像数据和面向对象的分类方法,提取出晚稻种植的变化区域。该文探索了变化强度计算和阈值确定的方法,并利用晚稻在不同时相的影像光谱特征变化来提取晚稻种植区域。试验结果表明:3种变化强度计算方法中,变化向量分析法对河流、滩地变化的抑制效果优于相关系数方法,而相关系数方法对山体阴影的抑制效果则优于变化向量分析法,向量相似度法对山体阴影非常敏感,对水田变化则敏感度较低;3种阈值确定方法中最小错误率方法比最大类间方差法更为精确,比双窗口变步长阈值搜寻法更为稳定。综合利用3种变化提取方法对平乐县晚稻种植面积变化进行检测,得到变化检测混淆矩阵总正确率为96.8%,稻田面积变化误差为2.85%。该方法可为作物种植面积的变化检测提供参考。Rice is the most important food crop in China and its production ranks first in three major grain crops which are wheat, rice and corn. Therefore, it is very important to understand planting area and growth of rice. Because change detection is time-sensitive, remote sensing data is often used as a data source for it. Most of widely-used change detection methods using TM image are at pixel level, or only use image of single time phase. Based on multi-temporal Landsat TM images, this research extracted change area of rice planting in Pingle County. The process explored the methods of calculating changes in intensity and determining the threshold. First, the paper selected images of heading stage and harvest stage as data sources, and except for crops, there are no significant changes of ground objects in a few months. Since the spectral characteristics of late rice in the two periods are different from those of other crops, the changes of the images at two stages in the same year could be used to extract the change area of rice planting. In the process, change vector analysis method, correlation coefficient method and vector similarity method were used to calculate the change intensity. Otsu method, minimum error rate method and the method based on double window with variable step size were used to determine the threshold of the change intensity map. The change area maps of rice planting extracted by nine combinations of the methods were compared. Comprehensive utilization of three methods for extracting changes could get change area of rice planting with higher accuracy. This paper chose minimum error rate method based on histogram curvature or the method based on double window with variable step size to determine the threshold. The intersection of change area based on change vector analysis method and change area based on correlation coefficient method could inhibit the pseudo change of river beach and mountain shadow. Then the rice area was extracted exactly according to the difference between NDVI values and th

关 键 词：遥感 作物 提取 变化向量 相关系数 面向对象 水稻 变化检测 

分 类 号：S127[农业科学—农业基础科学]