Rice Mapping in a Subtropical Hilly Region Based on Sentinel-1 Time Series Feature Analysis and the Dual Branch BiLSTM Model

Timely and accurate information on rice cultivation makes important contributions to the profound reform of global food agricultural system, promotes development sustainable agriculture. With all-day all-weather observing ability, synthetic aperture radar (SAR) can monitor distribution in tropical subtropical areas. To solve problem misclassification with no marked signal during flooding period hilly areas, this paper proposes a new feature combination dual branch bi-directional long short-term memory (DB-BiLSTM) model achieve high-precision mapping using Sentinel-1 time series data. Based field investigation data, backscatter curves area were analyzed, characteristic index (VV − VH)/(VV + VH) (VV: vertical emission receipt polarization, VH: horizontal polarization) for small areas land was proposed effectively distinguish non-rice crops period. The DB-BiLSTM designed, ensuring independent learning multiple features combining both VH features. city Shanwei, Guangdong Province, China, selected as study area. Experimental results showed that overall accuracy 97.29%, kappa coefficient reached 0.9424. Compared other methods, obtained by method maintained good integrity had less misclassification, which demonstrated method’s practical value effective tasks.