New Generation Indonesian Endemic Cattle Classification: MobileNetV2 and ResNet50

Abstract

Cattle are an essential source of animal food globally, and each country possesses unique endemic cattle races. However, categorizing cattle, especially in countries like Indonesia with a large cattle population, presents challenges due to costs and subjectivity when using human experts. This research utilizes Computer Vision (CV) for image data classification to address this urgent need for automatic categorization. The main objective of this study is to develop a mobile-friendly model using CV techniques that can accurately detect and classify Indonesian endemic cattle races, such as Limosin, Madura, Pegon, and Simental. To achieve this, an object localization approach is employed to extract multiple features from distinct regions of each cattle image, including the head, ear, horn, and muzzle areas. The authors evaluate two CV algorithms, ResNet50 and MobileNetV2, to assess their performance in cattle race classification. The dataset used is facial photos of 147 cows. The results indicate that ResNet50 outperforms MobileNetV2, achieving a training data accuracy of 83.33% compared to MobileNetV2's 77.08%. Moreover, the validation accuracy of ResNet50 (76.92%) significantly surpasses MobileNetV2's (38.46%). The novel contribution of this research lies in developing a cost-effective and efficient solution for identifying endemic cattle breeds in Indonesia. The mobile-friendly model based on ResNet50 demonstrates superior accuracy, enabling cattle farmers and researchers to categorize cattle races with higher precision, reducing manual effort, and minimizing costs. In conclusion, this research provides a valuable advancement in automatic cattle categorization using CV techniques. By offering a practical and accurate model that considers Indonesia's specific cattle breeding conditions, this study contributes to the sustainable management and conservation of endemic cattle races while enhancing the efficiency of cattle farming practices.