Non-invasive assessment of heat stress in cattle: an approach based on machine learning and infrared thermography

Author(s)

Alex Vinícius da Silva Rodrigues

• Document type: Master's Dissertation
• Press: Pirassununga.
• Institution: Universidade de São Paulo (USP). Faculdade de Zootecnica e Engenharia de Alimentos (FZE/BT)
• Defense date: 2022-02-21

Examining board members

Rafael Vieira de Sousa; Késia Oliveira da Silva Miranda; Daniella Jorge de Moura

• Advisor: Rafael Vieira de Sousa

Abstract

Recent studies in animal production have investigated technologies and computational models for predicting the level of thermal comfort through noninvasive and automatic measurements. To contribute to this theme, the project aims to build and test computational models for predicting the level of thermal comfort of dairy cattle using a database obtained by an experiment in a climatic chamber. The experiment was conducted for 45 days with 10 Holstein calves randomly distributed in two groups. The animals were placed in the climatic chamber (in a Tie Stall system) and exposed to two heat waves. During the experiment period, in addition to the meteorological data of the facilities, rectal temperature, respiratory rate and infrared thermography (IRT) data from different areas. of the body (eye, forehead, rib and flank) to extract temperatures values and the Thermal Signature (feature extraction method explored at work). For the modeling step, different algorithms based on Machine Learning (artificial neural networks, support vector machine, decision tree and closest k-neighbors) were evaluated using different combinations of inputs for classifying the thermal stress level attribute. To compare the classification models, the parameters obtained from the generated confusion matrix (accuracy, precision and recall) were used as a metric. The best results were obtained with the Random Forest and support vector machine algorithms. The Thermal Signature proved to be more efficient as a predictor attribute of the models when compared to the point temperatures extracted from the TIVs. Models with accuracies above 90% in the classification of animal heat stress level were obtained in this work. The results obtained evidence the potential of using machine learning associated with data extracted from infrared thermography for the classification of animal heat stress level. (AU)

• FAPESP's process: 19/26828-1 - Prediction of thermal comfort level for dairy cattle: method based on machine learning modelling using climate chamber experiment data
• Grantee: Alex Vinicius da Silva Rodrigues
• Support Opportunities: Scholarships in Brazil - Master