A Non-invasive Cattle Identification Using Muzzle Pattern Recognition in Livestock Management

In the Philippines, small hold or backyard cattle farms play a vital role in food security, contributing about 85% of beef production. However, issues such as theft, fraud, and ownership conflicts remain common, especially in decentralized cattle marketplaces. This project introduces a non-destructive cattle identification system that uses muzzle pattern recognition to enhance livestock management. A deep learning-based application was developed using TensorFlow, Keras, and OpenCV to identify cattle by analyzing their muzzle patterns, offering a nondestructive, non-invasive, tamper-proof, cost-effective, and secured livestock identification. The application is deployed as a web application which is hosted on Hostinger it was tested using five cattle in Sariaya, achieving a score of 91.69%, 92.61%, 93.62%, 87.75%, and 94.27%, additionally five cattle is from Pakistan from the open-source dataset which was able to have an output of 96.85%,97.22%,96.39%,94.81%, and 97.99% and was able to generate a 91% of model accuracy. User Acceptability is used to evaluate the system which is based on ISO/IEC 25010:2011 and was answered by nine respondents in where they were asked to use the application at least 3 times to test the application, and they are required to input new entries of cattle and identify individually. It was able to generate an overall weighted mean of 4.26 which resulted that the users strongly agree that the system was able to achieve functionality, reliability, and usability as a supportive tool for livestock management.