Illegal Construction Site Detection Using Deep Learning

Illegal construction activities pose significant challenges to urban planning, safety, and regulation enforcement. Traditional methods for detecting such activities often rely on manual inspections and are laborious and highly dependent on resources. In this study, a novel approach for illegal construction site detection using deep learning techniques is proposed. This research proposes the use of high-resolution satellite imagery combined with the potent capabilities of the YOLOv8 object detection model to identify and flag unauthorized construction endeavors. Cutting-edge deep learning technique was deployed in this work to integrate critical contextual data related to land use, ensuring a superior differentiation between legal and illicit construction sites. Pictures of different construction sites were collected across Mumbai using Google Earth, a threshold area of 20,000 square feet was set to determine the occupied area as legal, and area above this has been marked as illegal. 98.2% “Illegal” instances were identified while the model accuracy in correct identification of “Legal” instances was 76%. 70% “Background” instances were correctly identified. Through this advanced methodology, urban planners, policymakers, and enforcement agencies are endowed with a comprehensive and efficient tool, potentially revolutionizing the monitoring, regulation, and governance of urban infrastructural growth.