An Explainable AI-Based Model for Feature Identification from Drone Orthophotos

With the increasing adoption of drones for large-scale surveying and monitoring, the ability to automatically interpret orthophotos has become critical for efficient infrastructure management and rural development. Accurate analysis of drone orthophotos is vital for effective mapping, infrastructure assessment, and data-driven planning. This work presents an AI-driven method for automatic feature extraction and categorization of roof types from drone-captured orthophotos, utilizing datasets under the SVAMITVA Scheme. This focuses on the use of image analysis as a key element for achieving accuracy. Key features include building area, road length, vegetation index, perimeter, water area, and roof type. Image processing techniques such as HSV transformation, contour extraction, and Hough line detection enable effective segmentation of structures, vegetation, and roads. Roof type prediction is further enhanced with explainable AI techniques like LIME and SHAP, providing transparency and insights into model behaviour. The proposed pipeline combines feature engineering, SMOTE for class imbalance correction, Random Forest algorithm, and Gradient Boosting models for classification. This methodology offers a scalable and interpretable solution for drone-based infrastructure analysis, contributing to smarter and more accountable rural mapping initiatives.