PERFORMANCE ANALYSIS OF HSI-RGB FEATURE FUSION FOR FIRE BLIGHT DETECTION IN APPLE LEAVES

Abstract

This study evaluates machine learning models—Support Vector Machine, Random Forest, and XGBoost—for fire blight detection in apple leaves using hyperspectral (HSI) data and fused HSI-RGB features. Results show that while HSI data alone enables strong classification (F1-score up to 0.93), fusing HSI with RGB features significantly enhances performance. The Random Forest model with fused features achieved the highest accuracy and F1-score (0.98). Visual assessments further confirm improved localization of infected regions with feature fusion. These findings demonstrate that multimodal data integration and ensemble learning substantially advance early, accurate fire blight detection for precision agriculture.