Prediction of Grape Leaf Black Rot Damaged Surface Percentage Using Hybrid Linear Discriminant Analysis and Decision Tree

Vitis vinifera, the common grape vine is susceptible to Guignardia bidwellii fungus which causes black rot damage to its leaves, flowers, and eventually the fruit. Instances of subjective assessment by farmers have been observed to mistreat the specific fungus for isolation. To solve this problem, allied feature-based machine learning and computer vision were utilized in classifying healthy and diseased grape leaves and predicting the damage area percentage. The dataset with 943 images is composed of matured healthy and diseased grape leaves and distinctively obtained via digital camera. CIELab thresholding via lazy snapping segmented the region of interest for both healthy, diseased, and whole leaf regions. The information gain of the classification tree (CTree) was used as a determinant in selecting the most significant spectro-textural-morphological grape leaf signatures (A, S, G, Y, H, B, Cr, entropy, correlation, contrast, energy) concerning predicting damaged surface percentage. Linear discriminant analysis (LDA) produced a highly commended output of identifying grape leaf health conditions with an accuracy of 97.79% than CSVM, NB, and KNN. The regression tree (RTree) showed the best performance in predicting the fungal damage area with $\mathrm{R}^{2}$ of 0.943 than GPR, RSVM, and LDA. The application of seamless LDA-CTree-RTree to continuously classify if the grape leaf is healthy or diseased and quantify the infected area for crop phenotyping and plant disease resistance analysis is acceptable and effective.