Deep Learning-Based Piglet Crushing Detection on Edge Computing Platforms

In this study, we aim to demonstrate the effectiveness of integrating Artificial Intelligence of Things (AIoT) with deep learning to detect piglet crushing incidents by sows, a leading cause of piglet mortality on pig farms. We use a You Only Look Once (YOLO) based deep learning approach to detect piglet crushing events in the pig pen. Considering the limitations of edge AIoT computing platforms, which have limited computational power, the original YOLO model is optimized using TensorFlow Lite and TensorRT. We evaluate the performance of the lightweight YOLO-based piglet crush detection model on various single-board computers (SBCs), including the Jetson Nano, Raspberry Pi, and Jetson Orin Nano, to validate its real-world applicability and effectiveness. Initial performance evaluations focus on two key metrics: computation time per frame and cost efficiency. The results indicate that the Jetson Orin Nano achieves the lowest computation time, while the Jetson Nano provides the highest cost efficiency. We plan to extend the performance comparison to more resource-constrained IoT devices, such as those based on the ESP32.