Enhancing Elephant Emotion Detection Using YOLOv5, YOLOv8, and YOLOv9: A Study on Performance and Explainability

Elephant emotion detection has marked a crucial component owing to the increased casualties due to human-elephant conflicts. The difficulty has increased due to the lack of algorithms that quickly identify emotions with explainable features. This paper proposes a holistic architecture that detects the emotions of elephants using YOLOv5, YOLOv8, and YOLOv9 models, The study aims to integrate explainable AI features such as Eigen-CAM to manifest the training and testing capabilities of models. In addition, we evaluated the capability of the proposed system using labeled datasets that distinguish elephant emotions to angry, happy, and sad. Our experimental results, conducted at the IoT Cloud Research laboratory, revealed training accuracy of 79.3% for YOLOv5, 79.5% for YOLOv8, and 81.1 % for YOLOv9 and test accuracy of 93%, 95%, and 95% for YOLOv5, YOLOv8, and YOLOv9. Additionally, we recorded the impact of resolutions while performing model evaluations. The observations learned from the experiments will benefit AI-assisted machines to detect elephant emotions that can be utilized by mahouts or people living near forest locations.