A Deep Learning Biometric Cattle Identification System to Combat Cattle Rustling in Nigeria
Cattle rustling remains a serious threat to livestock farming, food security, food supply, and economic stability in Nigeria. Traditional cattle identification methods like ear tagging and branding are prone to damage and offer limited durability and accuracy. This study presents reliable and non-invasive cattle biometric identification system based on deep learning that utilizes muzzle-print images to address these limitations. Muzzle prints are similar to human fingerprints that are unique and tamper-proof. These make muzzle prints a viable biometric trait for cattle identification. A dataset comprised of 1,038 muzzle-print images from 70 cattle collected from two local farms and the Zenodo public repository was preprocessed through normalization, resizing and reduction of noise. Data augmentation techniques such as rotation, zooming, and brightness were used to increase dataset diversity. The ResNet-50 convolutional neural network was fine-tuned for muzzle print classification due to it known strong performance in image recognition. The model effectively identified cattle under varying environmental conditions to achieve an accuracy rate of 99.9%. Limited data availability, variability in image capture conditions and hardware constraints were some of the challenges encountered during the implementation stages.
Download Paper