Detection and Classification of Canopy of South India from Satellite Images Using YoloV5

This study presents an automated, large-scale approach for detecting and classifying canopy patches across South India using satellite imagery and advanced object detection techniques. Unlike localized studies, this work leverages 500 high-resolution images from Google Earth Engine and 4600 annotated canopy sections from Roboflow to ensure a diverse, well-structured dataset. A robust preprocessing pipeline, including data augmentation, scaling, and normalization, enhances model adaptability to varied terrains. A key novelty is the integration of attention mechanisms into YOLOv5, improving detection accuracy in dense vegetation, urban overlaps, and mixed landscapes. The model achieves 85% accuracy, surpassing traditional pixel-based methods with object-aware segmentation.