Enhanced Water-body Texture Extraction from GF-2 Satellite Imagery Using Transfer Learning Approach with Gray Level Difference Matrix

Accurate water-body monitoring through remote sensing image analysis is crucial for efficient management. This paper introduces a hybrid deep learning approach to overcome traditional methods' limitations of higher misclassification rates, time constraints, and model complexity. The focus is on distinguishing water-body image from non-water type by analyzing textural features. To achieve this, a novel methodology that combines features extracted using EfficientNet-B5 and Gray Level Co-occurrence Matrix (GLCM) is proposed. Integration of GLCM with deep learning classifiers provide comprehensive texture information and reduced sensitivity to illumination variations. Experimental results, conducted on MATLAB R2020a over GF-2 satellite images, showcase superior performance over ResNet-101 and VGG-16 networks. The proposed model demonstrates efficiency in computation, reduced complexity, and improved accuracy, thereby contributing significantly to advancements in water resource management.