Computer vision applied to the monitoring of individual body mass of finishing pigs

Abstract

Recent studies in animal production have shown that performance measures related to body mass help in animal management and in the quality of animal protein produced. In this context, the project seeks to build and evaluate a computer vision system based on a 3D camera and radio frequency identification (RFID) for automated monitoring of the performance of pigs in the finishing phase. For this purpose, experimental data will be used from 50 pigs housed in two pens between the 8th and 20th week, with body mass between 20 and 120 kg. Weighing data associated with images obtained with an RGB-D camera associated with RFID identification of each animal will be used. From the database, algorithms for extracting dimensional characteristics of the dorsal surface associated with computer models for body mass prediction will be evaluated. For the modeling stage, different algorithms based on Machine Learning will be evaluated (eg artificial neural networks, support vector machine, random forest and k-nearest neighbors) using different combinations of inputs for prediction. The tests to compare the prediction models will use linear regression parameters (correlation coefficient and residual) and error (percentage mean error and root mean square error) as analysis metric. Through the results of the project, it is sought to highlight not only the best computational techniques for building the models, but also to highlight the potential of using the RGB-D camera associated with RFID identification.