Cattle AutoID: Biometric for Cattle Identification: Cattle AutoID

Abstract

Existing solutions of animal identification (i.e., cattle in this research project) are based on RFID, ear tag, and microchip. However, they are facing with difficulties of high cost, dislodged and lost, and harm to human operators and animals. Therefore, this paper proposes a biometric based solution of cattle identification using cattle's face images. The proposed method is developed using a convolutional neural network (CNN) for both main steps of face localization and face recognition. The face localization model is trained using a Single-Shot Detector (SSD) architecture, where the face recognition model is trained based on FaceNet. The proposed method is validated using our dataset containing 2,432 cattle images from 152 different cattle. It achieves 94.74% and 83.45% for subject-based and image-based accuracies respectively.