多源遥感数据协同的地块级农作物精细提取

Fast and Precise extraction of field crops based on multi-source remote sensing data

关键词:  农作物分类  多源遥感数据  遥感协同应用  基于地块

Key words:  crop extraction  multi-source remote sensing data  remote sensing collaborative application  land parcels-based classification

摘要:

【目的】 针对云南省大理市耕地地块不规整、破碎且农作物空间种植结构复杂的特点，结合多源数据在时间和空间分辨率的优势，达到准确地提取农作物信息的目的。【方法】 协同BJ-2数据和Sentinal-2数据进行农作物精细信息提取。首先，利用空间分辨率较高的BJ-2数据进行面向对象的图像分割，获得农作物地块信息；其次，在农作物物候规律分析的基础上，通过标准差分析获得关键时相，利用相应时间分辨率较高的Sentinal-2数据获取农作物地类信息，实现基于地块的小春农作物的快速精细提取。【结果】 采用实地调查地块真值与提取地类生成混淆矩阵进行精度验证，总体精度和Kappa系数分别为87.4%和0.83。其中，连片种植的农作物如蚕豆和马铃薯提取精度较高，地块细碎且内部种植结构复杂的作物提取精度略低。【结论】 多源遥感数据协同的农作物提取方法，通过高分辨率影像上获得的对象分析单元能很好地对单一地块中的农作物空间特征进行统计分析，很大程度上弥补了中分辨率影像由于分辨率偏低所导致的混合像元处错分的不足；不仅能从耕地地块级别获得农作物种植结构，更直观地反映农作物种植，能有效提升农作物提取的精细化程度，有利于精细化的农作物种植结构管理。

Abstract:

[Purpose] Based on the characteristics of highly fragmented cultivated land and complex crop spatial planting structure in Dali City，combine the advantages of multi-source data in temporal resolution and spatial resolution to precisely extract crop information.[Method] BJ-2 with high spatial resolution and Sentinal-2 with high temporal resolution are collaborative applied to precisely extract crop information. Firstly，bJ-2 data with high spatial resolution was used for object oriented image segmentation to obtain crop plot information. On the basis of the analysis of crop phenology law，the key time phases were obtained by standard deviation analysis，and crop land category information was obtained by using sentinal-2 data with high corresponding time resolution，so as to realize the rapid and fine extraction of early spring crops based on plots.[Result] The accuracy was verified by using the truth value of the field survey plot and the confusion matrix generated by the extraction of land class. The overall accuracy and Kappa coefficient were 87.4% and 0.83 respectively. Among them，the extraction accuracy of crops planted continuously，such as broad bean and potato，is higher，while the extraction accuracy of crops with fragmented plots and complex internal planting structure is slightly lower.[Conclusion] The collaborative crop extraction method of multi-source remote sensing data can perform statistical analysis on the spatial characteristics of crops in a single plot through the object analysis unit obtained from the high-resolution image，which to a large extent makes up for the lack of misclassification at the mixed pixel caused by the low resolution of the medium resolution image. It can not only obtain crop planting structure from the level of cultivated land，but also reflect crop planting more intuitively，which can effectively improve the refinement of crop extraction and is conducive to the refinement of crop planting structure management.