Regresar

Android Mobile Application for Cattle Body Condition Score Using Convolutional Neural Networks

Abstract:

The livestock sector is the set of activities related to raising cattle to take advantage of reproduction, dairy production, and beef benefits. In this sector, a vital factor is a cattle’s body condition, considered a nutritional indicator since the subcutaneous body fat level found in certain anatomical points determines the animal’s thinness or fatness levels. Therefore, it is a clue in defining nutritional deficiencies, a common problem in the livestock industry. Providing a timely cattle body condition assessment may prevent nutritional issues that improve cattle’s health, reproduction processes, and dairy production. This study aims to develop an Android mobile app assessing Bos Taurus cattle body condition through computer-vision techniques and Deep Learning. The app was developed following the XP agile methodology and the Flutter, TensorFlow, and Keras frameworks. For this end, three CNN models were trained: Yolo, MobileNet, and VGG-16, for different tasks within the App. Models were evaluated using quantitative metrics such as Confusion Matrix, ROC Curve, CED Curve, and AUC. The ISO/IEC 25022 standard and USE questionnaire were used to assess the mobile app quality in use. The mobile app achieved an accuracy of 0.88 between the manual body condition score (BCS) and the one predicted. Results proved that this application enables anyone to adequately assess cattle’s body condition using a conventional mobile device, contributing to the innovation of the livestock sector.