A Chainsaw-Sound Recognition Model for Detecting Illegal Logging Activities in Forests

Illegal logging activities in Kenya results to an increase in carbon emissions, creating a need to detect and prevent illegal logging activities. This paper proposes the use of an internet-of-things (IoT) based architecture for detection of logging sounds by chainsaw and a machine learning (ML) technique to identify and classify the collected environmental sounds. The IoT architecture, based on Long-Range (LoRa) wireless technology, will include devices fitted with sound sensors that are strategically placed in an identified site within the forest. Sound signals will then be transmitted in real-time to a cloud-based platform for storage, and classification using a temporal frequency convolutional neural network (TFCNN) model. The TFCNN model will include an attention mechanism for recognition of different sounds by their distinct characteristics and a feature representation module to further distinguish chainsaw sounds from other environmental sounds. Open-source datasets such as ESC-50 and FSC22 will be considered in model training but the latter will be utilized more due to its overall focus on forest acoustics.