Computer Vision-Based Autonomous Drone Surveillance For Illegal Deforestation Detection In Protected Forests

Abstract

Illegal deforestation in protected forests poses significant threats to biodiversity, carbon stocks, and ecological balance. This research presents an autonomous drone surveillance system integrating computer vision and deep learning to detect and monitor illegal deforestation activities in near real-time. High-resolution UAV imagery, including RGB and multispectral data, was collected across multiple forest sites and annotated for clearings, tree stumps, vehicle tracks, and logging equipment. Four deep learning models—YOLOv5, Faster R-CNN, U-Net, and DeepLabv3+—were implemented for object detection and semantic segmentation. Performance evaluation demonstrated that Faster R-CNN and DeepLabv3+ achieved the highest F1-scores of 0.90–0.92 and IoU values of 0.82–0.83, while YOLOv5 enabled rapid inference with a latency of 0.9 seconds per alert, suitable for real-time onboard deployment. Segmentation models accurately delineated cleared areas, with DeepLabv3+ achieving an area estimation error as low as 5.2%. Comparative analysis with related work indicated superior accuracy, operational efficiency, and scalability for practical forest monitoring. The system effectively managed to join fast alerting, high accuracy and effectiveness of UAV coverage, which resulted in actionable intelligence by forest management and enforcement agencies. It is shown that deep-learning-driven autonomous UAV surveillance is an achievable, scalable, and effective solution to illegal deforestation.