Morphological Phenotyping for Cattle Breed Identification from UAVs Images Using Deep Convolutional Neural Networks (DCNNs)


Creative Commons License

Çakmakçı C., Turan M., Çakmakçı Y., Assıs Ferraz P., Akay M., Bülbüller F., ...More

6. International Food, Agriculture and Veterinary Sciences Congress, Ganca, Azerbaijan, 22 - 23 September 2023, pp.196-197

  • Publication Type: Conference Paper / Summary Text
  • City: Ganca
  • Country: Azerbaijan
  • Page Numbers: pp.196-197
  • Van Yüzüncü Yıl University Affiliated: Yes

Abstract

Unmanned aerial vehicles (UAVs) and deep learning enable non-invasive livestock monitoring with increased precision compared with traditional hands-on methods. The detailed morphological data from UAV images enable better characterization of breed-specific traits relative to sparse on-farm records, which supports genetic studies to identify trait-associated genomic variants and genomic selection to genetically improve breeds. The aim of this study was to train deep convolutional neural networks (DCNNs) to classify dairy cattle breeds based on visual characteristics, such as coat patterns, body shape, size, etc. A diverse set of RGB UAV images of Holstein, Simmental, and Brown Swiss breeds was collected. The images were preprocessed and segmented to contain a single cow each, and subsequently categorized into training, validation, and testing sets. To determine the most effective architecture for breed classification, a custom DCNN (C-DCNN) was compared with the pretrained Xception, VGG19, and ResNet50 models. Both the C-DCNN and Xception models emerged as superior choices for breed classification, demonstrating the feasibility of this approach. These results demonstrate the potential of integrating UAVs with DCNNs for robust, non-invasive cattle monitoring to support precision livestock farming. This methodology's adaptability suggests possible extensions to other livestock species. The promising outcomes in breed classification also indicate potential capabilities in other applications, such as early detection of lesions or wounds using UAV images, facilitating improved livestock health and welfare. Expanding dataset can further enhance DCNN model performance for broader precision livestock applications, such as breed classification, anomaly detection, and phenotypic characterization.

Keywords: Breed classification, Computer Vision, Dairy Cattle, Deep Learning, Drone, UAVs.