Single-Season Rice Mapping Method Based on Multi-Temporal Sentinel-1 Data and Attunet Model

Timely and accurate monitoring of rice cultivation areas is essential for ensuring national food security and achieving sustainability goals. The Synthetic Aperture Radar (SAR) technology enables all-day, all-weather Earth monitoring and is extensively applied in rice mapping studies. Most rice mapping methods based on SAR imagery rely on prior knowledge and empirical thresholds specific to certain regions. The accuracy of these mapping results is often limited by traditional machine learning algorithms, which fail to fully exploit the deep abstract features present in multi-temporal images. Furthermore, existing deep learning-based rice mapping methods tend to produce incomplete and misclassified results for some rice cultivation areas. To overcome these issues, this study has improved and innovated upon existing deep learning-based rice classification models. We propose a simple yet robust rice mapping method. Using this approach, we conducted rice mapping in the Suihua City of Heilongjiang Province in 2023. The results demonstrate that the proposed method has an overall accuracy of 97.51%, with mapping accuracy and user accuracy of 95.36% and 97.57%, respectively. This method guarantees complete classification results and minimizes misclassifications, providing a viable solution for achieving high-precision single-season rice mapping.