Long history paddy rice mapping across Northeast China with deep learning and annual result enhancement method-Training Dataset

We have generated the training and validation labels for the paddy rice mapping. Both the Landsat images and their corresponding masks were rotated by 5°. Subsequently, all the Landsat images and masks were cropped into several small images with dimensions of 256 × 256. These small images covered the Landsat images completely without any overlap, and any cropped image without paddy rice pixels was removed. The training and validation sets for Landsat 5 images were 29906 and 9968, respectively, the training and validation sets for Landsat 8/9 images were 50956 and 16985, respectively.