Adaptive Multi-Model AI Framework for Geospatial Feature Extraction from Drone Orthophotos

Abstract

This research attempts to tackle the problem of geospatial extraction and mapping through high-resolution orthophotos of drone aerial photography. Traditionally, manual extraction is quite tedious and laborious with inherent inconsistencies associated with it, especially when applied to wide-ranging rural landscapes with varying land patterns. Object variations like different sizes, textures, and appearances and also the presence of infrastructural assets in small numbers makes the process of traditional feature extraction difficult. In order to resolve the challenge of the existing techniques, the proposed method introduces an AI-based framework that utilizes modern deep learning techniques to automatically detect, segment, and classify geospatial features. This paper aims at building footprints, rooftops (RCC, tiled, and tin), roads, water bodies, and utility assets (transformers and overhead tanks).