Detection of Animals based on their Species by Implementing Bio-Acoustic Classification for Forest Wildlife Monitoring

Wildlife conservation requires an efficient and proper monitoring system. Traditional methods for tracking forest animals, such as Infrared Cameras, IoT networks, and drone technology, are no longer efficient in the current scenario. Present situation requires extensive equipment and are often costly. This research introduced bioacoustics combined with machine learning and deep learning techniques to classify animal sounds, offering a scalable solution for wildlife monitoring. Identifying animal species is crucial in detecting their presence and mapping their territories. The proposed model utilized sound event detection to identify animals based on their species. Our study focused on five animals—tigers, leopards, otters, elephants, and antagonists. We created a dataset of 3,851 audio clips from tigers, leopards, otters, elephants, and other forest area sounds. Augmentation techniques like noise injection and pitch shifting are implemented to enhance the dataset's quality. We trained and evaluated several models, including SVM, KNN, and a deep learning-based Customized CNN demonstrating the best performance, achieving an accuracy of 91%.