To support agricultural development in Thailand, accurate data collection and analysis of land use are essential. Understanding the spatial distribution and growth patterns of key crops enables better planning and resource allocation. This study proposes a deep learning-based approach for land cover classification, specifically targeting three significant crops: rice, rubber, and palm. A Convolutional Neural Network (CNN) is employed to classify satellite imagery into these three categories. The datasets used in this research are derived from high-resolution Pleiades satellite imagery and consist of three independent datasets, each containing 200,000 image tiles of 100x100 pixels. For each crop type, a dedicated CNN model are trained and optimized, achieving classification accuracies exceeding 90%. After prediction, a post-processing step is implemented to merge tile-level classifications into continuous land cover maps. This enables a clearer spatial visualization of crop distribution. Furthermore, a clustering algorithm is applied to identify individual agricultural fields, which facilitates further analysis. Vegetation health and maturity are assessed using the Normalized Difference Vegetation Index (NDVI), from which the approximate age of the crops is inferred. These parameters are then used to estimate the potential agricultural yield or production for each field. To validate the approach, a large area of 100 square kilometers is analyzed, and the model’s classification results are compared against manually labeled reference data provided by the Thailand Space Agency. The comparison reveals a classification discrepancy of -12% for palm crops and approximately -20% for both rice and rubber, demonstrating the model's high potential for scalable crop monitoring.