Preventing Illegal Deforestation using Acoustic Surveillance

Abstract

With the rapid increase in deforestation and the subsequent impact on global warming, rainforest protection is the ffrst step to preventing drastic climate change. Audio classiffcation based on audio recognition techniques is promising as they have consistently performed better than humans in urban sound classiffcation. The challenge arises as the research performed on the audio classiffcation of natural sounds such as the rainforest are in their preliminary stage and the shortage of a strongly labelled dataset. This paper proposes a solution to prevent illegal deforestation in rainforests with acoustic surveillance and deep learning. Further, this works to adopt transfer learning on three different models, YAMNet, AlexNet, and ResNet-50, to discover which methodology yields the most practical and effective approach to send real-time alerts for chainsaw incursions in rainforests. We also introduce an architecture that allows our solution to deploy over mobile phones. The investigated method is further extended in an automated prototype that future researchers can easily integrate into solutions based on cloud technology for real-world deployment.