Application of machine learning methods for the differentiation of fungal diseases in strawberry based on hyperspectral image analysis

Abstract

Fungal diseases have a significant negative impact on strawberry yield. Their detection and differentiation using hyperspectral measurements is a possible alternative to traditional methods. In this study, strawberry leaves infected with Ramularia Tulasnei, Marssonina potentillae and Dendrophoma obscurans with visible symptoms of the disease were used for hyperspectral analysis. The reflection spectrum of leaves was recorded with a Photonfocus hyperspectral camera (wavelength range 475–900 nm, 149 channels) under laboratory conditions using the line scanning method. This research has aimed to compare four machine learning methods: spectral angle mapper (SAM), support vector machine (SVM), k-nearest neighbors (KNN) and linear discriminant analysis (LDA). Classification models were built based on the full spectrum, as well as on 12 vegetation indices (VI) as spectral features. The results demonstrated that the SVM model based on full spectra reached highest classification accuracy 94%. The KNN model performed slightly worse with 91% accuracy. The performance of models based on VIs was lower than that of models based on full spectra with an accuracy range of 78–85%.