其他知识库

Human-elephant conflict (HEC) affects people and wild elephants negatively, and support for harmonious coexistence is needed. With the current human footprint, wildlife is displaced, and people living near wildlife want safe interactions. Conservation interventions are needed to manage human-elephant coexistence in realtime. This research, using deep learning models, provides the fundamental mechanics for acoustic detection of elephants in an automated early-warning system, currently under development. We examine the use of convolutional neural networks (CNNs) for classifying Asian elephant (Elephas maximus) sounds and non-elephant sounds. The results demonstrated the ability of CNNs to process bioacoustics data across various sample sizes, with the best-performing model achieving 98.45 % average test accuracy (balanced sample sizes, a k-fold approach with 10 % for testing). But when we infer CNN models built with Sri Lankas elephant vocalizations with unseen Malaysias elephant vocalizations, the performance of the models dropped to an average of 67.93 % accuracy and F1 score between 0.67 and 0.81, regardless of the initial training dataset size. We used Principal Component Analysis to compare 15 sound parameters extracted from spectrograms of elephant calls from Sri Lanka and Malaysia, and found that the sound characteristics between the two subspecies largely overlapped but with some differences. We conclude that the CNN models can detect elephant sounds but perform best with local data. The use of bioacoustic monitoring and automated detection can potentially support harmonious coexistence between humans and elephants, but for endangered species targeted by poachers, safeguards are needed. Additionally, we need discourse on research ethics and local communitys rights.