請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/89769
標題: | 基於人工智慧輔助建立豬隻健康與室內環境品質間相關性 Establishing Correlation Between Pig Health and Indoor Environment Quality Assisted by Artificial Intelligence |
作者: | 蔡艾倫 Ai-Luen Tsai |
指導教授: | 陳佳堃 Jia-Kun Chen |
關鍵字: | 精準畜禽飼養管理,影像辨識模型,豬隻疾病, precision livestock farming system(PLF),image recognition model,swine disease, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 本研究建立影像辨識方法協助豬舍管理者蒐集豬隻攝食量及排泄物的健康資料,再透過建立豬舍內部空氣品質與豬隻健康間的相關性以達到降低豬舍人力短缺及提供農民輔助監控豬隻健康的資訊。實際的場域量測於彰化縣一豬舍進行空氣品質監測母豬產房現場環境條件,包括溫度、濕度、PM2.5、PM10、CO2、TVOC、NH3。基於YOLOv5建立影像辨識模型進行收集母豬健康資料,包含手寫文字辨識模型及排泄物健康判斷模型,透過訓練模型獲得最合適的手寫文字辨識(model 13)及母豬排泄物(feces model 3)健康判斷模型,mAP_0.5及精準度(precision)皆達到0.99,手寫文字辨識模型之最終測試準確率達到97%。後續分析母豬健康狀態的資料與豬舍內部空氣品質間相關性可知,空氣品質指標除氨氣外,其餘皆對於母豬的健康狀態有顯著的影響,TVOC、CO2及溫度的增加可能會對母豬攝食量的健康狀態產生負面影響,而溫度、濕度、PM10的增加可能會對母豬的排泄物健康狀態產生負面影響,可提供在未來建置母豬疾病預測模型時衡量變項的依據。綜合上述研究結果可得知在建置影像辨識模型時的重要參考要項,訓練樣本和標記的質量是關鍵因素,而模型的參數調整和過擬合(overfitting)與否皆需特別注意。透過豬隻健康狀態與空氣品質間的相關性分析,可提供豬場管理者協助改善母豬的生產環境,降低母豬疾病風險及提高生產效益,並提供未來研究建置疾病預測模型的建置參考。 To address the labor shortage in agricultural farms and facilitate pig health monitoring, a field study was conducted in a farrowing house located in Changhua County. The study focused on monitoring various environmental parameters, including temperature, humidity, PM2.5, PM10, CO2, TVOC, and NH3. An image recognition model based on YOLOv5 was developed to gather sow health data, specifically for handwriting recognition and feces health assessment. Through rigorous model training, exceptional precision rates were achieved for both handwriting recognition (0.99) and sow feces health judgment (0.99), with a final test accuracy of 97% for the handwriting recognition model. Furthermore, the study conducted an analysis to examine the correlation between sow health status and air quality indicators. The results indicated significant impacts of TVOC, CO2, temperature, humidity, and PM10 on sow health, excluding ammonia. Elevated levels of TVOC, CO2, and temperature were found to adversely affect sow intake health status, whereas increased temperature, humidity, and PM10 had negative effects on sow excretion health status. These findings serve as a fundamental basis for the development of future sow disease prediction models. The construction of an accurate image recognition model necessitates high-quality training samples and meticulous attention to model parameters to mitigate overfitting. By investigating the relationship between pig health and air quality, it is possible to optimize sow production environments, reduce disease risks, enhance production efficiency, and establish a valuable reference for future research endeavors. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/89769 |
DOI: | 10.6342/NTU202301085 |
全文授權: | 未授權 |
顯示於系所單位: | 環境與職業健康科學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-111-2.pdf 目前未授權公開取用 | 9.44 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。