利用深度卷積神經網路於雞隻活動力與溫溼度指數之監測

Monitoring chicken activity and temperature-humidity index using deep convolutional neural networks

林宸儀(Chen-Yi Lin)

指導教授 ： 郭彥甫

國立臺灣大學/生物資源暨農學院/生物產業機電工程學研究所/碩士(2018年)

https://doi.org/10.6342/NTU201801694

摘要

家禽和雞蛋是飲食中的主要蛋白質來源，不但營養價值高，也是許多加工食品的主要原料。此外，家禽年產值占臺灣畜牧業總銷售量的31.19%，是國內很重要的經濟來源。在熱帶和亞熱帶地區，熱緊迫是家禽產業面臨的挑戰性問題之一。熱緊迫的影響不僅降低雞隻的生長速度和雞蛋的品質, 甚至導致死亡率的提升。典型的雞隻熱緊迫症狀包括飲水頻率的增加與進食量的減少，藉由上述行為以保持體溫的恆定。在雞隻管理上，如果能及早發現熱緊迫現象，絕對是穩定家禽產業生產的關鍵。傳統的雞隻監控以溫濕度指數 (THI) 作為熱緊迫現象之評估。然而, THI只是一個間接指標。實際上熱緊迫的現象可能會因為雞隻品種和飲食供應而異。本研究直接利用影像和深度學習演算法對雞隻的行為進行監測。在此研究中，利用樹莓派和一個網路攝影機來獲取雞舍內的雞隻圖像，並建立了深度卷積神經網路分類器作為個別雞隻偵測之模型。最後，將個別雞隻偵測結果做活動力追蹤。此系統能即時且有效地監控雞隻活動力和環境因子，並將雞隻活動力結合環境因子做分析，可在未來做為評估雞隻熱緊迫狀態之參考。

關鍵字

雞隻活動力 ； 深度學習 ； 嵌入式系統 ； 快速區域卷積神經網路 ； 熱緊迫 ； 溫溼度指數

並列摘要

Poultry and eggs are major sources of dietary protein. Their production accounted for at least 30% of total animal husbandry sales in Taiwan. As in the tropical and subtropical areas, heat stress is one of the most challenging problems for the poultry industry in Taiwan. Heat stress reduces growth rate of chickens and quality of eggs and is even associated with sudden deaths. Typical heat stress symptoms include increasing the frequency of drinking, reducing feed intake and movement to maintain a constant body temperature. Conventionally, the level of heat stress was indicated using temperature-humidity index (THI). THI is, however, an indirect indicator. The level of heat stress may vary with chicken varieties and dietary supplies to the chickens. This work proposed to monitor behaviors of chickens directly using time lapsed images and deep learning algorithms. In the study, an experimental coop was constructed to host ten chickens. An embedded system was designed to acquire images of the chickens at a rate of one frame per second and to measure the temperature and humidity of the coop. A Faster R-CNN model was then developed on a personal computer to identify and localize the chickens in the images. The movements and drinking frequencies of the chickens under various THI values were then analyzed. The proposed method achieved a mAP of 93.0% and an overall accuracy of 98.16% in chicken identification, and achieved an accuracy of 98.94% in chicken tracking.

並列關鍵字

Chicken activities ； deep learning ； embedded system ； faster region-based convolutional neural network (Faster R-CNN) ； heat stress ； temperature-humidity index (THI)

參考文獻

Baweja, H. S., Parhar, T., Mirbod, O., & Nuske, S. (2018). StalkNet: A Deep Learning Pipeline for High-Throughput Measurement of Plant Stalk Count and Stalk Width. In Field and Service Robotics (pp. 271-284). Springer, Cham.

Boykov, Y., Veksler, O., & Zabih, R. (2001). Fast approximate energy minimization via graph cuts. IEEE Transactions on pattern analysis and machine intelligence, 23(11), 1222-1239.

Cheema, G. S., & Anand, S. (2017, September). Automatic detection and recognition of individuals in patterned species. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases (pp. 27-38). Springer, Cham.

Chen, G., Han, T. X., He, Z., Kays, R., & Forrester, T. (2014, October). Deep convolutional neural network based species recognition for wild animal monitoring. In Image Processing (ICIP), 2014 IEEE International Conference on (pp. 858-862). IEEE.

Council of Agriculture, Executive Yuan, Taiwan. (2015). Results of animal husbandry statistics survey. Taipei, Taiwan: Taiwan Agricultural Research Institute, Council of Agriculture, Executive Yuan.