Presymptomatic Detection of Fire Blight in Apple Orchards Using Portable Diffuse Reflectance Spectroscopy: A Machine Learning Approach

Fire blight, caused by the bacterium Erwinia amylovora, poses a severe threat to apple orchards worldwide, leading to significant production losses and potential tree death. Traditional detection methods are labor-intensive and often imprecise, while molecular techniques, though accurate, require specialized equipment and expertise. This study explores the use of smartphone-assisted spectroscopy combined with machine learning for the early detection of fire blight in apple orchards. Utilizing the K-means clustering algorithm, three distinct treatment groups were identified. Subsequent classification using random forest models achieved high accuracy, 86.89% for the control group, 89.71% for the antibiotic group, and 90.35% for the non-antibiotic group. Feature importance analysis highlighted key spectral wavelengths crucial for the infection status classification. The results demonstrate the potential of spectroscopic techniques for rapid, noninvasive detection of fire blight, suggesting a promising tool for enhancing disease management in agriculture. Future research should focus on optimizing chemical treatments, analyzing data point relationships, and comparing various machine learning models to further improve detection accuracy.