Detection of Deforestation by Mining and Agricultural Activities Using Convolutional Neural Networks

Rimachi V.W.S.; Quispe M.V.; Aragon-Navarrete R.N.; Ulloa-Gallardo N.J.; Holgado-Apaza L.A. · RISTI - Revista Iberica de Sistemas e Tecnologias de Informacao (2024)

Resumen

Deforestation, which represents a global challenge and a threat to biodiversity and environmental balance, is aggravated by activities such as illegal mining. This study proposes a novel solution using Convolutional Neural Networks (CNNs). These models, which include VGG16, MobileNet and VGG19, together with a custom architecture model, are used to analyze satellite images and classify areas affected by mining and agriculture. Transfer learning techniques are implemented using Google Earth imagery. The results reveal that VGG16 exhibits the most outstanding performance in terms of key metrics, with an accuracy of 56%, a sensitivity of 55% and an F1 score of 0.56. This finding highlights the effectiveness of RNCs, especially VGG16, in identifying areas deforested by human activities. This opens opportunities for monitoring and mitigating environmental impacts, which is crucial for addressing deforestation and its consequences.

Instituciones: Universidad Nacional Amazónica de Madre de Dios