Deep Learning-Based Spatial Pattern Modeling for Land Use and Land Cover Classification Using Satellite Imagery

Accurate classification of Land Use and Land Cover (LULC) is crucial in Remote-Sensing (RS) and satellite imaging to understand Earth's surface attributes. However, existing methods often face challenges in effectively extracting and categorizing complex spatial patterns from satellite imagery. The evolution of deep learning techniques has offered promising advancements in this domain, yet further enhancements are needed to achieve optimal performance. This study introduces a novel deep learning-based spatial pattern modeling technique designed to address these challenges. The proposed method leverages the Inception-V3 model to extract detailed features from the EuroSAT dataset comprising 27,000 images across 10 LULC classifications. By fine-tuning hyperparameters and conducting rigorous training-validation experiments, the model achieves notable performance metrics: an accuracy of 0.9943 and a validation accuracy of 0.9850, with corresponding losses of 0.0184 and 0.0566. This approach represents a significant advancement over traditional methods, offering enhanced accuracy and efficiency in LULC classification, thereby facilitating more informed decision-making in environmental monitoring and spatial analysis.