Piglet Teat Counting Based on Multi-Object Tracking

The mammary traits of sows are crucial for breeding selection, as they affect milk production and piglet survival rates. Manual counting of sows' teats is inefficient and often affected by challenges like occlusion, stains, and uneven lighting. Since the number and relative positions of piglet teats remain stable as they grow, this study proposes an automatic counting method using multi-object tracking. Abdominal videos are captured on a piglet care platform, and a lightweight object detection network is employed to detect teats and count them. The Teat-Track algorithm is developed to improve counting accuracy by tracking detection boxes across multiple frames, reducing errors caused by occlusion or motion blur. The results show that the proposed method achieves 92.2% accuracy, with an average counting time of 2.43 seconds (ranging from 0.8 to 4.6 seconds), demonstrating high real-time capability and accuracy.