Trees have Ears: An Acoustic Surveillance and TinyML-Based for Detecting Illegal Logging

This study presents an innovative acoustic surveillance system utilizing TinyML to detect illegal logging in the rainforests of the Philippines. To overcome the limitations of traditional methods, the researchers integrated acoustic data capture with machine learning algorithms on the RP2040 microcontroller. Two feature extraction methods—spectrogram and Mel Filterbank Energy (MFE)—were compared for training machine learning models in both binary and multiclass sound classification tasks. The spectrogram-based binary model demonstrated exceptional accuracy, efficient resource utilization, and fast execution, making it ideal for resourceconstrained environments. Meanwhile, the MFE binary model excelled in real-world acoustic analysis and performed well in multiclass classification, achieving 84.4% accuracy at an 8 kHz sampling rate and 89.6% at 16 kHz. Furthermore, tests confirmed reliable chainsaw sound detection up to 20 meters away. These findings offer an effective solution for illegal logging detection and provide valuable insights into optimizing models based on context and hardware configurations. This system contributes significantly to global efforts in ecosystem monitoring and conservation.