Extraction of winter wheat planting area across multiple phenological periods based on feature optimization using Sentinel-1/2 data

Winter wheat constitutes a fundamental cereal crop in China's agricultural system, playing a pivotal role in national food security. Timely and accurate acquisition of winter wheat cultivation area distribution is crucial for effective management, yield estimation, and ensuring food security. This study focuses on Hebi City as the research area, selecting Sentinel-1 and Sentinel-2 imageries from October 2021 to June 2022. The research was conducted on the Google Earth Engine (GEE) cloud computing platform, employing a multi-feature approach that integrated polarization characteristics, spectral properties, vegetation indices, textural features, and topographic parameters across various phenological stages of winter wheat. The random forest algorithm was implemented for crop classification and area extraction. The results show that: (1) The optimized feature sets constructed based on the Pearson correlation coefficient can improve overall classification accuracy, with an overall accuracy exceeding 90% across all schemes. (2) Adding both texture and polarization features can improve the overall classification accuracy of the heading stage and the full phenological period, most significantly in the full phenological period; (3) The extraction scheme for winter wheat planting area during the milk ripening stage, considering the optimized texture features and preferred polarization features, is the most effective method, achieving an overall accuracy of 98.1% and a Kappa coefficient of 0.976. The achievements of this research have broad application prospects in guiding regional precision winter wheat cultivation, optimizing agricultural resource allocation, supporting grain yield prediction, and ensuring national food security. It is expected to provide strong data support and technical references for agricultural management departments in making scientific decisions.