Morphological Phenotyping for Cattle Breeds Classification From Unmanned Aerial Vehicle Imagery via Computer Vision and Deep Learning

Abstract

Advancements in unmanned aerial vehicle (UAV) technologies have facilitated a novel approach to dairy cattle breed morphological identification. The objective of this study was to employ UAV images, analyzed through deep convolutional neural networks (DCNN), to classify dairy cow breeds. The dataset comprises of 2004 RGB UAV images of dairy cows, including Holstein, Simmental, and Brown-Swiss breeds, obtained from the cattle breeding facility at Van Yüzüncü Yıl University. The images were preprocessed and segmented to contain a single cow each, and subsequently categorized as training (70%), validation (20%), and testing (10%) datasets. To determine the most effective architecture for breed classification, we compared a custom DCNN (C-DCNN) model to well-established pre-trained models including Xception, VGG19, and ResNet50. The C-DCNN demonstrated remarkable performance, achieving precision, recall, accuracy, and F1 scores of 0.98. Among the pre-trained models, Xception demonstrated superior results, with perfect accuracy and an F1 score of 1.00. Conversely, the VGG19 model exhibited a higher level of accuracy; nevertheless, it exhibited lower precision, recall, and F1 scores when evaluated on the test set, compared to the C-DCNN and Xception models. In contrast, ResNet50 displayed the lowest level of performance, with an accuracy of 0.74 and the highest levels of loss. This study demonstrates the potential of integrating DCNN models with UAV technology in precision livestock farming, offering a robust and efficient system for cattle breed classification.