The livestock of Pakistan includes different animal breeds utilized for milk farming and exporting worldwide. Buffalo have a high milk production rate, and Pakistan is the third-largest milk-producing country, and its production is increasing over time. Hence, it is essential to recognize the best Buffalo breed for a high milk- and meat yield to meet the world’s demands and breed production. Pakistan has the second-largest number of buffalos among countries worldwide, where the Neli-Ravi breed is the most common. The extensive demand for Neli and Ravi breeds resulted in the new cross-breed “Neli-Ravi” in the 1960s. Identifying and segregating the Neli-Ravi breed from other buffalo breeds is the most crucial concern for Pakistan’s dairy-production centers. Therefore, the automatic detection and classification of buffalo breeds are required. In this research, a computer-vision-based recognition framework is proposed to identify and classify the Neli-Ravi breed from other buffalo breeds. The proposed framework employs self-activated-based improved convolutional neural networks (CNN) combined with self-transfer learning. Moreover, feature maps extracted from CNN are further transferred to obtain rich feature vectors. Different machine learning (Ml) classifiers are adopted to classify the feature vectors. The proposed framework is evaluated on two buffalo breeds, namely, Neli-Ravi and Khundi, and one additional target class contains different buffalo breeds collectively called Mix. The proposed research achieves a maximum of 93% accuracy using SVM and more than 85% accuracy employing recent variants.
Read full abstract