Optimized Deep-Learning-Based Method for Cattle Udder Traits Classification

We propose optimized deep learning (DL) models for automatic analysis of udder conformation traits cattle. One the is represented by supernumerary teats that in excess normal number teats. Supernumerary are most common congenital heritable Therefore, major advantage our proposed method its capability to automatically select relevant images and thereafter perform teat classification when limited data available. For this purpose, we experimental on image dataset collected using a handheld device consisting combined depth RGB camera. To disclose underlying characteristics data, consider uniform manifold approximation projection (UMAP) technique. Furthermore, comprehensive evaluation, explore impact different augmentation techniques performances DL models. also only combination integrate three channels with channel generate four data. present results all terms performance metrics, namely accuracy, F-score, precision, sensitivity. The reveal higher level improves approximately 10%. Our outperforms reference methods recently introduced literature.