Identification of the Citrus Greening Disease Using Spectral and Textural Features Based on Hyperspectral Imaging

Area of Science:

• Agricultural remote sensing
• Plant pathology
• Image analysis

Background:

• Citrus greening disease (Huanglongbing or HLB) poses a significant threat to citrus production worldwide.
• Early detection of HLB is crucial for effective disease management and preventing widespread crop loss.
• Current detection methods can be time-consuming and may not identify the disease in its initial stages.

Purpose of the Study:

• To investigate the application of spectral and textural features from hyperspectral imaging for early-stage identification of citrus greening disease (HLB).
• To evaluate the performance of classification models using reduced spectral data and image textural features.
• To assess the feasibility of developing a portable detection system for HLB.

Main Methods:

• Acquisition of 176 hyperspectral images of citrus leaves (healthy, HLB-infected, zinc-deficient).
• Extraction of spectral data and application of Principal Component Analysis (PCA) and Successive Projection Analysis (SPA) for dimension reduction.
• Development of classification models using Least Square-Support Vector Machine (LS-SVM) with spectral and textural features.
• Extraction of textural features using gray-level histogram and Gray-Level Co-occurrence Matrix (GLCM) from selected wavelengths.

Main Results:

• Classification models using PCA and SPA selected variables achieved average accuracies of 89.7% and 87.4%, respectively.
• Incorporating textural features from selected wavelengths significantly improved classification accuracy, reaching 100% for three-class samples.
• SPA-ranked textural features demonstrated high efficacy in distinguishing between healthy, HLB-infected, and zinc-deficient citrus leaves.

Conclusions:

• Spectral and textural features derived from hyperspectral imaging are effective for early-stage HLB detection.
• Image textural features, particularly when combined with specific wavelengths, offer a highly accurate method for identifying HLB.
• The study provides a foundation for developing portable, multispectral imaging systems for rapid and reliable HLB diagnosis in the field.