Abstract
This paper proposes an AI-powered framework for illegal logging detection using CNNs, satellite imagery and geospatial analytics.
Keywords
Full Text
Forest degradation continues to increase globally despite advances in environmental monitoring. Traditional field monitoring remains expensive, slow and difficult to scale across the vast forested regions most at risk of illegal logging.
Artificial intelligence provides a scalable mechanism capable of analysing thousands of satellite images automatically, flagging canopy loss patterns consistent with illegal extraction long before ground teams could detect them manually.
The proposed architecture employs a Convolutional Neural Network trained using Sentinel-2 satellite imagery combined with transfer learning techniques. Two model variants were evaluated: a baseline CNN and a ResNet50 backbone, both fine-tuned on a labelled dataset of confirmed logging sites across three West African forest reserves.
The ResNet50 variant achieved 98.5% accuracy, 98.0% precision and 97.9% recall on the held-out test set, outperforming the baseline CNN's 97.8% accuracy. These results suggest the approach is viable for near-real-time monitoring deployments where field verification resources are limited.
References
- Goodfellow I., Bengio Y., Courville A. Deep Learning. MIT Press, 2016.
- FAO. Global Forest Resources Assessment 2025. Food and Agriculture Organization of the United Nations.
- United Nations. Sustainable Development Goals Report 2025.