End-to-End Individual Pig Detection Based on Transfer Learning

Nowadays, monitoring animals manually on a large scale is challenging, while external devices can provide more prosperous and accurate information. Therefore, there are already applications incorporating wearable sensors to monitor individual pigs. With continued breakthroughs in computer vision, low-cost visible light cameras can help monitor pigs. CNN-based object detection methods have been applied to individual pig detection, but most rely on hand-designed components and require prior knowledge. Based on previous researchers' progress, we propose using DETR, an end-to-end object detection method, to detect individual pigs. Also, to adapt to our natural pig farm environment, we trained the model with transfer learning to achieve an AP of 0.995 on our collected dataset.