Deep Learning Techniques for Beef Cattle Body Weight Prediction
Document Type
Conference Proceeding
Date of Original Version
7-1-2020
Abstract
Following the weight of beef cattle is of great importance to the producer. The activities of nutrition, management, genetics, health and environment can benefit from the weight control of these animals. We explore different deep learning models performance in the regression task of predicting cattle weight. This is a hard problem since moving from 3-D space to 2-D images presents a loss of information in object shape, making weight prediction more difficult. A model that produces good results in this problem could potentially be applied more abstractly to similar problem spaces. We analyzed convolutional neural networks, RNN/CNN networks, Recurrent Attention Models, and Recurrent Attention Models with Convolutional Neural Networks, and show that convolutional neural networks achieve the highest performance. Our top model averages a MAE of 23.19 kg. This is nearly half the error as previous top linear regression models which reached an error of 38.46 kg.
Publication Title, e.g., Journal
Proceedings of the International Joint Conference on Neural Networks
Citation/Publisher Attribution
Gjergji, Mikel, Vanessa De Moraes Weber, Luiz Otávio Campos Silva, Rodrigo Da Costa Gomes, Thiago L. De Araújo, Hemerson Pistori, and Marco Alvarez. "Deep Learning Techniques for Beef Cattle Body Weight Prediction." Proceedings of the International Joint Conference on Neural Networks (2020). doi: 10.1109/IJCNN48605.2020.9207624.