Po-Cheng HsiehEn-Chung LinYan-Fu Kuo2024-10-282024-10-282024https://www.scopus.com/record/display.uri?eid=2-s2.0-85206114028&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/722511Pork plays an essential role in diet worldwide as it is one primary source of dietary protein. According to Food and Agriculture Organization, pork accounts for 94% of the increased demand in global meat production growth. As the global population rises rapidly, it is challenging to meet the demand in pork while ensuring high quality using conventional pig farming approaches. Pig production primarily comprises three stages, namely farrowing, nursery, and fattening. In the first week of the nursery stage, piglets from multiple farrowing crates are put together. The piglets are fragile because they are just weaned. In addition, piglets from different farrowing crates have various strength. Strong piglets may bull weak piglets. Thus, in conventional pig farming practices, farmers need to frequently patrol the nursery houses, manually observe piglet conditions, and troubleshoot irregular situations. However, manual observation is time-consuming and may not detect the irregular situation early enough. Thus, this study proposed to monitor key behaviors and movement of weaned piglets in the first week for the nursery stage using convolutional neural networks (CNNs). The key behaviors include feeding, drinking, and aggression. Two models were used. A Yolov7 combined with SORT tracking algorithm was trained as the pig detection model (PDM) to both localize and track individual piglet and detect interactive piglets in nursery. An EfficientNet combined with LSTM was trained as the pig behavior recognition model (PBRM) to recognize piglet behaviors. The PDM achieved a mean average precision of 94.5% in pig detection. The PBRM achieved an accuracy of 86.5% in pig behavior recognition. This research can help farmers to find the abnormal crates and improve the farm management. Moreover, it may raise the income of pig farm by increasing the breeding rate of piglets and reducing staffing costs.falseconvolutional neural networksdeep learningfarm managementPig behavior detection systempig behaviors[SDGs]SDG2conference paper10.13031/aim.2024010532-s2.0-85206114028