2020-01-012024-05-14https://scholars.lib.ntu.edu.tw/handle/123456789/659693摘要：豬隻養殖佔據國內畜牧業產值中相當重要的比例，且在所有發育階段中，仔豬相對較弱需要更多關注。造成死亡率提升如仔豬互相咬尾造成感染、分娩後五天內被母豬壓死等，是主要兩項需要被監測的異常行為。傳統上針對仔豬行為異常之觀察常依賴人力，然而在廣大的豬舍中也無從得知個別豬隻之狀況，此外近年豬隻養殖管理也面臨人力不足的問題。因此本計畫提出一套利用嵌入式系統以及深度學習的自動化監測系統，其中嵌入式系統包括樹莓派、相機及麥克風。接著，藉由嵌入式系統所錄製的影片用來訓練深度學習模型。經過訓練後，模型可用來自動解釋仔豬活動的行為，所以此方法能有效地減少豬舍管理時間上的成本。<br> Abstract: Pig husbandry accounts for a considerable proportion in the domestic livestock production. Among all the development stages, piglets are relatively vulnerable and need more attention. Abnormal activities of piglets, such as tail biting and sow crushing, are two major causes of high mortality. These abnormal behaviors need to be monitored. Conventionally, the observation of abnormal behaviors are performed manually. However, manual observation is time consuming. Also, pig husbandry is facing the issue of labor shortage. Therefore, this project proposes to automatically monitor piglet activities using embedded systems and deep learning. The embedded systems comprise Raspberry Pi, cameras, and microphones. The videos recorded by the embedded systems are used to train deep convolutional neural network models. Once trained, the models can be used to interpret piglet activities automatically. Hence, the proposed method can effectively reduce the time needed for pig house management.深度學習物件偵測仔豬行為Deep learningObject detectionPiglet activity