High-value fruit biometric identification via triplet-loss technique

This thesis proposes a novel method for biometric authentication of fruits based on their distinctive rind patterns, similar to fingerprint identification. Luxury fruits, highly valued in Japan, currently rely on serial numbers, QR codes, and RFID tags for authentication, which can be forged or replicated. By implementing biometric authentication using rind patterns, the trust and value of these fruits can be significantly enhanced, while also preventing fraud and counterfeiting in the agricultural industry. The study introduces a melon identification system that utilizes a convolutional neural network (CNN) with a triplet loss function, enabling accurate identification even with variations in lighting, shadows, and angle. The proposed method overcomes the limitations of previous approaches by capturing important features through CNN’s Automatic feature identification. This research contributes to the field of agricultural product authentication, providing a secure and reliable method that can be extended to other products, increasing customer trust and market value.

本论文提出了一种基于水果独特果皮纹路的生物特征认证新方法，其原理与指纹识别类似。目前在日本市场备受青睐的高端水果，其认证多依赖序列号、二维码与射频识别（Radio Frequency Identification，RFID）标签，但此类方式极易被伪造或仿制。通过采用果皮纹路开展生物特征认证，可显著提升此类高端水果的品牌信任度与市场价值，同时有效遏制农业领域的造假与仿冒行为。本研究搭建了一款甜瓜识别系统，该系统采用搭载三元组损失函数的卷积神经网络（Convolutional Neural Network，CNN），能够在光照、阴影及拍摄角度存在变化的场景下实现精准识别。所提方法通过卷积神经网络的自动特征识别能力提取关键特征，从而克服了过往方法的局限性。本研究为农产品认证领域提供了一套安全可靠的解决方案，该方法可推广至其他品类产品，进而提升消费者信任度与产品市场价值。