Classification of Paddy Rice Planting Area Through Feature Selection Method Using Sentinel-1/2 Time Series Images

Utilizing remote sensing technology to accurately and efficiently extract paddy rice planting area is crucial for ensuring food security. In southern Jiangsu, cloudy and rainy weather impairs the effectiveness of optical satellite images, while complex surface coverage reduces the precision of paddy rice classification. Therefore, this study took Liyang City as the study area, reconstructed Sentinel-2 cloud-free time series optical images, and extracted spectral features, vegetation indexes, and other features, in combination with the polarization features of the Sentinel-1 time series radar images. The optimal feature subset was selected through the feature selection method, and machine learning algorithms were optimized for paddy rice planting area classification. Results indicated that: 1) The reconstruction of cloud-free time series images with the Cloud Score+ method and the integrated NSPI and MNSPI approach was stable and effective, with correlation coefficients (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">r</i>) exceeding 0.87 and low values for indicators such as root mean square error (RMSE), Robert's edge (Edges), and local binary patterns (LBP), meeting the requirements for paddy rice classification. 2) The classification accuracy of combining Sentinel-1 polarization features with Sentinel-2 spectral features could improve by up to 10.52% compared to before the combination. The combination of polarization features, spectral features, and difference features achieved the highest overall accuracy (OA), but the mapping exhibited salt-and-pepper noise. 3) The integration of multi-source remote sensing data and feature selection effectively improved paddy rice classification accuracy. The correlation-based feature selection and greedy step wise algorithms performed the best, with an OA of 93.97% and a Kappa coefficient (Kappa) of 0.9176, producing less mapping noise and higher classification accuracy for paddy rice. The study provides methodological support and a practical case for paddy rice planting area classification in the southern region using remote sensing.