Deep Learning Approach for Feed Intake Monitoring in Cattle

A crucial indicator of animal health and feed efficiency is the precise recording of animals’ daily feed intake. In the past, average feed intake for a bunch of animals or manual measures have been used, leading to human error and generally inconsistent results for individual animals. The aim of this study is to develop an accurate model by utilizing transfer learning to improve performance and save computational costs when classifying images of cattle feed according to the amount of leftovers. The suggested method classifies feedlot images into six bunk score categories (score 0, score ½, score 1, score 2, score 3, and score 4) using pre-trained models such as ResNet50, MobileNetV2, DenseNet121, EfficientNetB2, and EfficientNetB3. Among these the EfficientNetB3 model demonstrated superior performance, achieving the highest validation accuracy of 86%, thereby outperforming the other four pre-trained models utilized in this study.