Abstract:It is a challenge to acquire accurately regional crop structure information by using remote sensing technology at county scale for the possible reasons of cultivated land fragmentation, scattered distribution and complex planting structure. Jingtai County was taken as the research area, and multitemporal Sentinel-2A remote sensing image was used as the data source to construct the time sequences of five kinds of feature parameters, which were normalized difference vegetation index (NDVI), red edge normalized vegetation index (RENDVI), and their combinations (NDVI+RENDVI, NDVI-RENDVI as well as NDVI&RENDVI). The random forest method was used to classify the crops based on five kinds of feature parameters. The results were as follows: according to the shape, the multitemporal VI (vegetation index) feature curve of crops was divided into three types, which were called highlevel, including corn, rice, flax and potato, lowlevel, including onion, greenhouse crops and sandyfield crops, and openend type, including spring wheat and spring wheatautumn oil sunflowers, respectively. Openend type could be identified by images of May or September, meanwhile, highlevel type and lowlevel type could be distinguished by images of July or August. Among each type, crops could be identified by using images of different times. For highlevel type, four crops showed significant differences in the images of mature period, for lowlevel type, images in September could supply much information to distinguish two crops, and as for as openend type, there were significant differences for three crops all through four growing stages. The sequence of overall accuracy of classification results by five kinds of feature parameters from large to small was NDVI&RENDVI, NDVI, NDVI+RENDVI, RENDVI and NDVI-RENDVI.
