To categorize remote sensing images from a dataset with ten classes—'harbor', 'chaparral', 'tennis_court', 'industrial_area', 'parking_lot', 'forest', 'beach', 'overpass', 'airplane', and 'baseball_diamond'—the proposed model introduces a tailored deep CNN architecture. This custom model is designed to outperform conventional pre-trained CNN models and has been rigorously evaluated by comparing its performance against five different pre-trained models. The NWPU ESISC45 dataset was used for both training and evaluation. The effectiveness of the proposed CNN architecture is demonstrated by its performance on the NWPU-RESISC45 dataset. The model achieved an impressive accuracy of 95.5%, which is a strong indicator of its classification capabilities. Additionally, the model's precision and recall values were both 96%, reflecting its ability to correctly identify and retrieve relevant data from the images. Furthermore, the proposed model delivered exceptional results in additional performance metrics. The F1-score, which balances precision and recall, reached 96%, and the Intersection over Union (IoU), a metric for measuring the overlap between predicted and actual objects, was 91.4%. These outcomes highlight the model's robust performance in remote sensing image classification tasks.