整合熱影像與雷達呼吸監測系統於乳牛熱緊迫程度之分析

Abstract

乳牛的產乳量與其健康狀況密切相關。牠們的體溫將是反映身體狀況重要指標之一。紅外線熱影像技術已經有不少相關研究與應用，證實可以在非接觸式的情況下量測乳牛溫度並進一步研究與分析，因此將成為建立乳牛健康管理的自動化監測系統之關鍵。儘管有幾項研究報導了通過手持式熱影像攝影機測量乳牛溫度的方法，但手動測量耗時費力，在牧場的實際應用上並不實際。為了解決這些問題，本研究提出了一種自動化的非接觸式熱影像監測系統，可以有效地從即時熱影像監測系統擷取乳牛眼睛的溫度測量值。該系統利用深度學習方法進行泌乳牛眼睛偵測與定位。應用於即時乳牛眼睛偵測的YOLOv4模型經過訓練和優化；該模型之偵測率達到 0.99 及F1 score 達到0.99。從熱影像串流中擷取出許多包含乳牛眼睛的影像，再應用進一步的影像處理算法計算平均溫度。透過這種方法，熱影像攝影機對每頭經過前方的乳牛進行多次溫度測量。該系統安裝在台大動科系實驗牧場，並進行了長期實驗量測以紀錄個別乳牛溫度的變化，數據分析結果發現環境溫度、乳牛眼睛與熱影像攝影機間的距離及角度、個體牛隻差異、眼睛型態及牛隻行為都會對溫度測量都有很強的影響，代表眼睛溫度測量需要用環境溫度進行校正，並且需要對測量的溫度進行後處理以提高其準確性。此溫度監測系統也結合非接觸式雷達呼吸頻率系統進一步分析乳牛熱緊迫的程度。從實驗中分析群體牛隻結果可以得知，乳牛眼睛溫度與溫溼度指數之線性回歸斜率0.07，相關係數為0.69，乳牛呼吸頻率與溫溼度指數之相關係數為0.74，眼睛溫度與呼吸頻率之相關係數為0.60。以單一溫濕度指數做熱緊迫區分時，從眼睛溫度分析之溫濕度數值為71.8，以呼吸頻率分析之溫濕度數值為68.1。此結果驗證紅外線熱影像可以被運用於牧場中自動監測乳牛之健康狀態，也代表本研究所建立之系統在檢測乳牛發燒或評估熱緊迫方面具有未來應用之價值。

Dairy cows' milk production is closely related to their health status. One of the indicators reflecting their health status is their body temperature. Infrared thermal imaging has been demonstrated to process a high potential for non-contact measurement of dairy cow body temperature, which is crucial for establishing an automated health monitoring system for dairy cow management. Although several studies have reported on the dairy cow temperature measurement by handheld thermal imaging cameras, manual measurement is not a feasible approach for practical application in the dairy industry as it is laborious and time-consuming. To solve these problems, this work proposes an automated non-contact thermal imaging monitoring system that can efficiently take dairy cow eye temperature measurement from thermal video stream in real time. The system utilizes a deep learning approach for dairy cow eye detection. A YOLOv4 model for real-time dairy cow eye detection was trained and optimized; it yielded a hit rate of 0.99 and an F1-score of 0.99. For each detected sub-image containing the dairy cow eye in the video stream, a further image processing algorithm was applied to determine the mean temperature with its variance. With this approach, multiple temperature measurements are taken from each dairy cow walking by the thermal camera. The system was installed in the university experimental dairy farm and long-term experiments were carried out to assess the variations of temperature measurement. It was found that both the ambient temperature, the thermal camera distance and angle, individual differences, eye stage and cow behavior have strong effect on the temperature measurement, indicating that the eye temperature measurement needs to be corrected with the ambient temperature and measured temperatures need to be preprocessed in order to increase its accuracy. The experimental results also show that the proposed system has potential in regard to detecting dairy cow fever or assessing of heat stress. This system also combines the non-contact respiratory rate system to measure the respiratory rate of dairy cows. From the analysis of the results of the group cows in the experiment, it can be known that the linear regression slope between the eye temperature and the THI of dairy cows is 0.07, and the correlation coefficient is 0.69. The correlation coefficient between the respiratory frequency of dairy cows and THI is 0.74. The correlation coefficient between the eye temperature and the respiration rate is 0.60. When using a single THI for heat stress distinction, the THI value analyzed by eye temperature is 71.8, and the THI value analyzed by respiration rate is 68.1. This result validates that the thermal infrared camera can automatically monitor the health status of dairy cows and also represents that the system established in this study has future application value in detecting fever in dairy cows or assessing heat stress.