Two-Stepwise Hierarchical Adaptive Threshold Method for Automatic Rapeseed Mapping over Jiangsu Using Harmonized Landsat/Sentinel-2

Rapeseed distribution mapping is a crucial issue for food and oil security, entertainment, tourism development. Previous studies have used various remote sensing approaches to map rapeseed. However, the time-consuming labor-intensive sample data in these supervised classification methods greatly limit development of large-scale rapeseed studies. Regarding threshold methods, some empirical thresholding still need select optimal value, their accuracies decrease when fixed applied complex diverse environments. This study first developed Normalized Difference Index (NDRI), defined as difference green short-wave infrared bands divided by sum, find suitable feature distinguish from other types crops. Next, two-stepwise hierarchical adaptive (THAT) algorithm requiring no training was automatically extract Xinghua. Finally, two standalone Otsu with Canny Edge Detection (OCED) were across Jiangsu province. The results show that (1) NDRI can separate vegetation well; (2) OCED-THAT method accurately an overall accuracy (OA) 0.9559 Kappa coefficient 0.8569, it performed better than Otsu-THAT method; (3) had lower but acceptable Random Forest (OA = 0.9806 0.9391). indicates THAT model promising automatic