Detecting of Tree Cutting Sound in the Forest by Machine Learning Intelligence

Automated detection of certain acoustic signals in environments is becoming an emergent field. One such area explored is the detection of logging of trees in forests. In forests, tree cutting activities are illegal but due to a shortage of manpower and other resources, governments are not very successful in curbing this menace. In this research, we implemented CNN and Bi-CRNN models to detect the tree cutting sounds in the forest. For this research six classes dataset prepared, apart from that five classes dataset collected from Google Audio Set framework and ax cutting class samples are recorded in the real time forest environment. For the specific purpose of tree logging detection, the feature extraction techniques used include MFCC, Spectral Contrast, Mel-Spectrogram, Chroma, and Tonnetz. The model will be trained by splitting the feature extracted data into training and testing, where the training data size will be around 80% and testing data size will be around 20%. The training data will be used to train the model, whereas the testing data will be used to test the model. Finally, the classification has performed and validated with CNN and Bi-CRNN models to detect the sound in the forest environment. In these two executions models, the Bi-CRNN provided better 95.2% accuracy.