High-Value Fruit Biometric Identification via Triplet-Loss Technique

Abstract

High-valued products require an authentication method to provide the customers with a way to verify the product's genuineness. As the product's authenticity dramatically increases its value, the authentication method has to be reliable and secure. High-valued melon, a popular gift in Japan, is a suitable product for applying rind pattern identification, providing a new means of authenticity verification. As opposed to using only QR-code or RF-ID tags, implementing a rind pattern recognition can provide a more secure authentication method, improving customers' trust and further increasing the product value. A previous study on melon identification was done using a well-known method of fingerprint recognition known as minutiae feature extraction on the melon rind pattern. The study has shown accurate results in the controlled image acquisition environment. In this work, a method of melon identity matching is introduced by incorporating triplet loss function on a convolutional neural network, providing a system that can reliably match each melon image even with variating lighting, shadows, and angle.