2020-01-012024-05-14https://scholars.lib.ntu.edu.tw/handle/123456789/659798摘要：家禽是國內主要的肉類來源之一。在傳統管理方式中，業者多透過人力親自巡視雞舍以觀察雞隻是否出現異常行為。然而此方式不但費時、管理效率低，且家禽業正面臨人力短缺的問題。本計畫將建立固定式俯視影像系統以觀察雞隻生長與活動情形，並開發移動式近距離影像系統搭載於子計畫四之巡航機器人，進行近距離雞隻生長與活動情形觀察。此俯視影像系統將由相機陣列、深度學習演算法及加裝圖形處理器之伺服器組成，其中相機陣列將使用嵌入式系統及相機組成並用來擷取雞隻影像，擷取後的影像將被上傳到伺服器以訓練深度學習演算法，辨識異常的雞隻生長狀況。而移動式近距離影像系統則安裝於子計畫四之巡航機器人，進行近距離觀察雞隻生長與活動情況。當俯視影像系統發現有異常雞隻行為出現時，將通知搭配的機器人移動至該異常雞隻位置，並使用搭載的移動式影像系統近距離觀察雞隻狀況。本計畫期望透過俯視影像系統與移動式影像系統輔助雞舍管理。<br> Abstract: Poultry is a major source of dietary protein in Taiwan. Conventionally, chicken farmers patrol in chicken houses to manually observe abnormal behaviors of the chickens and to stimulate the chickens to move. However, manual approaches are laborious and time consuming, and poultry husbandry is facing the issue of manpower shortage. This project proposes to establish a top-view and a close-up monitoring system for observing chicken growth and activities. The top-view monitoring system will be composed of an imaging array, deep learning algorithms, and GPU servers. The imaging array will comprise several sets of embedded systems and cameras that are used to acquire videos or images of chickens. The acquired videos or images will be sent to the GPU servers, and then will be used to develop deep learning algorithms for detecting chickens of abnormal activities. The close-up monitoring system will be installed on the robots developed by project 4. The system will be composed of a camera, deep learning algorithms, and a GPU. Once the imaging array of the top-view monitoring system detects chickens of abnormal activities, the locations of the chickens will be sent to the robot. The robot will approach to observe the chickens in a close proximity. The developed monitoring systems are expected to help the monitoring of chicken houses.深度學習自動感測雞隻行為移動式影像系統Deep learningAutomatic detectionChicken activityMobile image system