Feature Fusion for Advancing Land-Use Scene Classification

Land-Use Scene Classification (LUSC) plays a crucial role in remote sensing, contributing to urban planning, environmental monitoring, and disaster management. This study addresses the challenges of traditional classification techniques by integrating three feature extraction methods: VGG16, Local Binary Patterns (LBP), and Gray-Level Co-occurrence Matrix (GLCM). VGG16 captures high-level spatial features, LBP focuses on local texture patterns, and GLCM enhances spatial texture analysis. The combination of these techniques offers a robust and comprehensive approach to land-use classification. Using the UC Merced Land Use Dataset, which includes 2,100 high-resolution images across 21 categories, the proposed methodology is tested and evaluated. Experimental results demonstrate that the fusion of VGG16 and LBP features yields the best performance, achieving 96% accuracy. The study highlights the importance of texture features, particularly LBP, in distinguishing visually similar categories, and explores the role of spatial features (GLCM) in identifying natural landscapes. The findings underscore the potential of hybrid feature fusion approaches in enhancing classification accuracy and robustness in real-world land-use analysis.