A LDA-based segmentation model for classifying pixels in crop diseased images

Effective segmentation of symptoms from crop diseased images is a vital important step in the timely detection of crop disease based on image processing techniques. Many of the formerly proposed methods still did not show a satisfactory performance in the extraction of symptoms from RGB images, especially when the images contain specularly reflected and shadowed parts. In this paper, we propose a novel approach to classify individual pixels in crop diseased images taken in the field as diseased or healthy. The approach is based on the machine learning algorithm linear discriminant analysis (LDA) and color transformation. Five color spaces were applied and compared over diseased images infected by four diseases commonly observed in cucumber crops - target spot, angular leaf spot, downy mildew and powdery mildew. The experimental results demonstrated that our proposed approach under RGB color space outperformed the other three contrast methods particularly for the images including shadowed and specularly reflected parts. Overall, the proposed LDA-based segmentation model can be used to the symptoms segmentation effectively.