Obtaining crop types and crop spatial distribution rapidly and accurately has great significance for water consumption prediction, planting structure adjustment, and precision agricultural management. Located in Southern China, the Dengta Basin Irrigation District is the main grain production area of Guangdong Province. The main crops in the Dengta Basin are paddy rice, peanut, and vegetable. The satellite imagery and programming code from the Google Earth Engine (GEE) cloud-based platform allows for crop mapping within specific timescales. However, few studies focus on crop types and land use/land cover (LULC) classification simultaneously in the agricultural land fragmentation area. Here, we proposed a method for crop types mapping based on Sentinel-2 imagery and the GEE platform. First, we used GEE to produce the annual composite image to execute LULC classification with the random forest classifier and obtained the agricultural land area. Second, we smoothed the composite Nominalized Difference Vegetation Index (NDVI) curves by Savitzky-Golay (S-G) filter and selected NDVI maximum feature threshold at the typical time phase for crops. Third, we used the NDVI threshold and the decision tree model to distinguish crop types. Finally, we generated the crop types map in Dengta Basin coupling with the LULC classification results. For the evaluation of crop classification, the overall accuracy was 87%, with a kappa coefficient of 0.79, and the proportion of crop planting areas is close to the official statistics. This research indicates that the proposed approach is promising for rapid, automatic, accurate, and cost-effective crop types mapping.