A method for detecting citrus huanglongbing based on hyperspectral imaging technology

By constructing a three-dimensional spectral database and a multi-scale hierarchical Transformer model, the problem of insufficient accuracy of hyperspectral imaging technology in cross-variety identification in the detection of Huanglongbing in citrus was solved, and high-precision detection of unknown varieties was achieved.

CN122368599APending Publication Date: 2026-07-10CHINA JILIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHINA JILIANG UNIV
Filing Date
2026-04-14
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Existing hyperspectral imaging technology suffers from a significant decline in classification performance and insufficient accuracy in cross-variety identification scenarios for citrus Huanglongbing detection, and cannot adapt to the spectral feature variations of unknown varieties.

Method used

The method employs a three-dimensional spectral database, ROI three-zone sampling, isolated forest algorithm for data cleaning, Savitzky-Golay filter and standard normal transform preprocessing, combined with continuous projection algorithm and competitive adaptive reweighted sampling algorithm to screen feature bands, and constructs a multi-scale hierarchical Transformer model for detection.

Benefits of technology

It significantly improves the accuracy of citrus Huanglongbing detection, especially showing an improvement of 6% to 8% for unknown varieties, and provides a high-precision non-destructive testing solution.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122368599A_ABST
    Figure CN122368599A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of precision agriculture technology, specifically disclosing a method for detecting Huanglongbing (HLB) in multiple citrus varieties based on hyperspectral imaging technology. The method first acquires hyperspectral images of healthy and diseased citrus leaves from multiple varieties, extracting spectral data by dividing regions of interest. Next, it uses an isolated forest algorithm for data cleaning and performs data preprocessing using a Savitzky-Golay filter and standard normal transformation. Then, it selects key feature wavelengths to achieve data dimensionality reduction. Finally, it constructs and trains a multi-scale hierarchical Transformer model for HLB classification across varieties, comparing its performance with that of a basic Transformer model. The constructed model shows improvements across all varieties. This invention effectively solves the problem of insufficient generalization ability of traditional hyperspectral detection models in multi-variety scenarios, significantly improving the accuracy and robustness of disease identification, and providing a feasible technical solution for non-destructive, high-throughput detection of HLB in citrus.
Need to check novelty before this filing date? Find Prior Art