Application of the convolutional network Mask R-CNN for the estimation of the body weight of the guinea pig

Abstract

Artificial intelligence can contribute in tracking the productive cycle of the cuy through the application of convolutional networks, being a necessity the estimation of its weight. This study focused on the application of the Mask R-CNN convolutional network, using a mobile application as a tool for image capture. The methodology covered the following stages: i) bibliographic review, ii) data collection (images and pig weights), iii) image processing through data augmentation, iv) construction of a dataset (image selection and data transformation);, v) adaptation and training of the convolutional network, vi) analysis of the results to validate its performance, and finally, vii) implementation of a mobile application as a weight estimation tool. A set of 6244 pig images with their respective weights and masks was managed to be collected, together with the Mask R-CNN network adaptation. These tasks led to a correlation of R2 = 80.2% with the validation set, as well as to the development of a functional prototype capable of estimating the weight of pigs using the camera of a cell phone.