An accurate identification method for Striped Red Fuji apples based on Gaussian-SIFT for cloud manufacturing

Modern agriculture is increasingly embracing cloud manufacturing and service-oriented models, creating demand for traceability technologies that are non-invasive, shareable, and accessible on demand. Accurate identification of each agricultural product at the individual level is a key link in traceability, ensuring reliable and continuous information throughout the entire supply chain. Conventional identifiers such as barcodes, Radio Frequency Identification (RFID), and laser etching can be tampered with, may require specialized devices, or can damage the product, limiting unified cloud management and cross-process collaboration. This study addresses individual-level recognition of Striped Red Fuji apples for unlabeled, low-cost, and non-invasive applications. We propose Gaussian-SIFT, an enhanced SIFT (Scale Invariant Feature Transform)-based method that enhances subtle skin textures through dual Gaussian filtering, optimizes feature detection and matching parameters, and introduces a template-based matching-rate evaluation to ensure stable feature extraction and matching. Experiments conducted across multiple devices, backgrounds, angles, storage periods, and large-scale samples demonstrate that Gaussian-SIFT improves cross-device matching by approximately 9% and angle robustness by 10.6% compared with traditional SIFT. The method also maintains high self-matching accuracy after 50 days of storage and achieves 100% recognition for identical objects in large-scale testing. These findings demonstrate that Gaussian-SIFT provides a reliable, label-free identification approach suitable for cloud service deployment, enabling integrated traceability throughout agricultural production, storage, and distribution processes.