Automated Detection of Tree-Cutting Activities in Forest Environments using Convolutional Neural Networks and Advanced Audio Feature Extraction Techniques

Detecting tree cutting is essential in preventing illegal logging and also conserving the forest’s natural resources. This research aimed to present a solution for illegal logging by developing a sound detection system capable of identifying tree-cutting activities in forest environments. Based on concepts of machine learning and deep learning techniques, this research introduced a customized Convolutional Neural Network (CNN) model that is used to classify different sound events related to tree cutting. The model is trained using a dataset of four classes, and each class of 1200 samples are chainsaws, axe cutting, and handsaw sounds, along with other sounds of the forest environment. In this model features extracted from sound samples such as MFCC, Mel-Spectrogram, and Tonnetz are to enhance the model’s classification accuracy. The proposed model with customized CNN architecture provided the classification accuracy of $\mathbf{8 2 \%}$ over the 4800 sample of the dataset.