Synergy of High-Resolution Optical and Time Series SAR Data for Extracting Crop Planting Structure

The crop planting structure refers to the spatial distribution and composition of crops in a region, which holds significant importance for agricultural production planning and services. In this study, focusing on Rudong County in Nantong City, a methodology is proposed that combines high-resolution optical and time-series SAR data. The approach employed 2020 Beijing-2 imagery along with Sentinel-1 time-series data. HRNet was utilized for the extraction of cultivated land parcels in the study area, and an LSTM model was employed to classify crops like wheat, rapeseed, rice, and maize within these parcels based on temporal features. This enabled the extraction of the crop planting structure information within the research area. The findings indicated that high-resolution optical imagery exceled in surface feature identification, and HRNet, using deep learning, accurately extracted cultivated land parcels. The LSTM model effectively leveraged polarimetric features from different stages of crops within parcels, demonstrating excellent performance in extracting crop planting structure, with an overall accuracy of 89.93%. Through the spatiotemporal synergy of high-resolution optical imagery and multitemporal SAR data, coupled with deep learning techniques, this study accurately extracted internal crop planting structure information within the research area. This methodology can serve as a basis for agricultural production management and adjustments to planting patterns at a county lever.