Abstract

Cattle are the cornerstone of dairy sector and are appropriately referred to as the most precious animal by small-scale farmers. Each cattle species is known for its special abilities and have different values in the market. Among cattle, when we consider buffalo breeds, most of them are black in color with similar structural makeup despite minimal visual differences. Hence, proposing an automated process to identify the type of buffalo breeds would be very helpful for first time dairy farmers to know the right breed and buy the right breed. In this research, by using Convolutional Neural Networks (CNN), and depth-wise convolutions, two models that classify five different buffalo breeds using machine knowledge have been proposed. The data was trained on three different pre-trained models (EfficientNet-b0, Visual Geometry Group 19 (VGG19), and MobileNet-V2) using a transfer learning approach. In addition, two CNN architectures are proposed to classify the buffalo breeds (CNN and DwiseNet). After comparing and analyzing the results of the proposed methodology, it was found that EfficientNet-b0 with 99% of accuracy and DwiseNet with 98% of accuracy would detect the buffalo breeds automatically in an efficient manner.

Key words: Convolutional Neural Networks, Cattle breeds, Deep Learning, Depth wise Separable Convolutions, Transfer Learning