Automatic prediction of stress in piglets (Sus Scrofa) using infrared skin temperature

• Published in: Computers and electronics in agriculture Vol. 168; p. 105148
• Main Authors: da Fonseca, Felipe Napolitano, Abe, Jair Minoro, de Alencar Nääs, Irenilza, da Silva Cordeiro, Alexandra Ferreira, do Amaral, Fábio Vieira, Ungaro, Henry Costa
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01-01-2020; Elsevier BV
• Subjects: Accuracy; Aerial thermography; Air temperature; Data mining; Developing countries; Ethical standards; Infrared cameras; Infrared imaging; LDCs; Machine learning; Meat; Model development; Pain; Paraconsistent logic; Pig production; Pork; Predictions; Product safety; Sex; Skin temperature; Temperature; Pig production; Model development; Machine learning; Paraconsistent logic
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2019.105148

•A machine learning algorithm was developed for detecting stress in piglets.•Paraconsistent logic was applied to extract the uncertainty of data.•Thermal image analysis detects stressful conditions in piglets.•The model detects thirst and cold stress conditions with high accuracy.•Infrared skin temperature is a non-invasive method to predict stress.
Pork consumption grows about 5% per year in developing countries. Ensuring food safety within ethical standards of meat production is a growing consumer’ demand. The present study aimed to develop a model to predict stress in piglets based on the infrared skin temperature (IST) using machine learning and the paraconsistent logic. A total of 72 piglets (32 males and 40 females) from 1 to 52 days old had the infrared skin temperature recorded during the farrowing and nursery phases under different stress conditions (pain, cold/heat, hunger, and thirst). The assessment of the thermal images was done using an infrared thermography camera. Thermograms were taken at ambient air temperatures ranging from 24 to 30 °C. The minimum infrared skin temperature (IST min) and the maximum infrared skin temperature (ISTmax) and the piglet sex were used as attributes to find the stress conditions (target). The attributes considered in the analysis were classified using the data mining method. The imaging technique is subject to certain contradictions and uncertainties that require mathematical modeling. The paraconsistent logic was applied to extract the contradiction from the data. The stress condition that had higher accuracy in the detection was that predicted by the cold (100%) using the ISTmin, and ISTmin plus the piglet sex, and thirst (91%) using ISTmax and ISTmax plus the piglet sex. The highest prediction of hunger was found using ISTmin (86%). Although the model was precise in detecting those stresses, the other stressful conditions in piglets such as pain that had an accuracy equal or less than 50%. Results indicate a promising assessment of stress condition in piglets using infrared skin temperature. We suggest the inclusion of other attributes in the machine learning process to amplify the use of the model.