<p>Soil moisture (SM) regulates the water cycle and energy balance between the surface and the atmosphere by influencing evaporation, runoff, storage, and seepage in the water cycle, and is an important indicator of dry and wet conditions of terrestrial soils, which is of great importance in the fields of agricultural production, water resource management, and ecological environmental protection. With the continuous development of remote sensing technology through the Cyclone Global Navigation Satellite System (CYGNSS) mission, the Global Navigation Satellite System Reflectometer (GNSS-R) method has been widely used to assess soil moisture. A large number of previous studies have only investigated the singular scalar attributes of delayed Doppler maps (DDM) of CYGNSS products, including DDM peaks and skewness, while neglecting the comprehensive exploitation of complete DDM images. In this paper, a convolutional neural network (CNN)-based model is used to elucidate the complex connection between DDM plot features and reflectance measurements, and the enhanced 36&#xa0;km × 36&#xa0;km resolution SM product in the Soil Moisture Active-Passive (SMAP) task is used as the target for training and post-prediction evaluation. This study shows that for test sets with different division ratios, the smaller the division ratio, the higher the prediction accuracy, with an R-value of 0.942 and an RMSE of 0.053 m³/m³. Based on this approach, cluster analyses based on a variety of selected Globcover parameters were performed, resulting in predictions that generally outperformed the training results for the entire dataset, with overall correlation coefficients of 0.96. The method improves the accuracy of soil moisture retrieval for specific Globcover land cover types, which provides an important basis for the rational allocation and scheduling of water resources and guarantees the sustainable use of water resources.</p>