Discrimination method and discrimination system for fermentation degree of tea pile based on near infrared spectrum

CN122113000AActive Publication Date: 2026-05-29CENT SOUTH UNIV +2
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CENT SOUTH UNIV
Filing Date
2026-04-24
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

In existing technologies, the methods for judging the degree of fermentation of tea rely on human experience, which is highly subjective and lacks a unified quantitative standard. Furthermore, single near-infrared spectroscopy or machine vision methods are not robust to light fluctuations and humidity differences, resulting in low judgment accuracy.

Method used

A fusion method based on near-infrared spectroscopy and image features is adopted. Contribution weights are calculated through multi-scale feature enhancement and gating mechanism. Near-infrared reflectance spectral data and tea surface image data are dynamically weighted and fused to form joint features, thereby improving the discrimination accuracy.

Benefits of technology

It has achieved standardized and online detection of the degree of tea fermentation, improved the accuracy and stability of the judgment, and can adaptively adjust the modal dependence under different environmental conditions to provide a reliable judgment of the degree of fermentation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122113000A_ABST
    Figure CN122113000A_ABST
Patent Text Reader

Abstract

The application discloses a tea pile fermentation degree discrimination method and system based on near-infrared spectroscopy. The method extracts first deep features from first data, performs multi-scale feature enhancement on the first deep features, and obtains first enhanced features. Second deep features are extracted from second data, and multi-scale feature enhancement is performed on the second deep features to obtain second enhanced features. First contribution weights of the first enhanced features and second contribution weights of the second enhanced features are calculated. Weighted fusion is performed according to the first enhanced features and the first contribution weights, and the second enhanced features and the second contribution weights to obtain fusion features. The first enhanced features, the second enhanced features and the fusion features are combined in residual to obtain joint features. The pile fermentation degree discrimination result of the tea to be discriminated is obtained according to the joint features. The method can fuse spectral and image features, dynamically perform weighted fusion and retain single-modal information, thereby improving the discrimination accuracy of the pile fermentation degree.
Need to check novelty before this filing date? Find Prior Art