Comparison of visible-near infrared and mid-infrared spectroscopy for classification of Huanglongbing and citrus canker infected leaves
Keywords:
Disease detection, classification, quadratic discriminant analysis, k-nearest neighbor, USA.Abstract
In this study, visible-near infrared spectroscopy and mid-infrared spectroscopy were compared to evaluate their applicability in classifying citrus leaves infected with canker and HLB from healthy citrus leaves. The visible-near infrared spectra in the range 350-2,500 nm and mid-infrared spectra in the range of 5.15-10.72 µm were collected from healthy and diseased (canker, HLB) leaves. Following the spectral data collection, the data were preprocessed and classification was performed using two classifiers, quadratic discriminant analysis (QDA) and k-nearest neighbor (kNN). The classifiers (QDA, kNN) resulted in an average overall and individual class classification accuracy of about 90% or more. Mid-infrared spectroscopy provided high classification accuracy especially in identifying HLB-infected leaves; while, visible-near infrared spectroscopy was better suited for canker detection. Both methods have their own merits such as visible-near infrared spectroscopy offers non-invasive disease detection; while mid-infrared spectroscopy represents the chemical profile of the leaf structure, which may allow potential detection in asymptomatic stages.
Keywords: disease detection, classification, quadratic discriminant analysis, k-nearest neighbor