Severity Assessment of Potato Leaf Disease Induced by Alternaria solani Fungus Using Hybrid Decision Tree and Support Vector Machine Regression

Alternaria solani is a destructive fungus affecting potato crops that promotes early and late blight diseases. Early detection of its manifestation is an imperative step to prevent its spread. However, manual inspection of leaves is vulnerable to inefficient and inconsistent monitoring. As a solution, the application of computational intelligence and computer vision in identifying healthy and damaged leaves and detecting the percentage of damaged area (PDA) by Alternaria solani is presented in this paper. A total of 552 image sets (200 late blight, 200 early blight, and 152 healthy leaves) were utilized. Vegetation segmentation was employed via lazy snapping and CIELab color space for the healthy regions and PDA. Spectral (RGB, HSV, L*a*b*, YCbCr), Haralick textural (entropy, correlation, contrast, homogeneity, energy), and phenotypic (leaf canopy area) were extracted from the lettuce leaf canopies. With the use of the decision tree (DT), this 18-feature vector was narrowed down to the 10 most significant features (R, G, S, a*, Cb, Cr, contrast, energy, entropy, leaf canopy area). Support vector machine (SVM) has the best performance (which has 100% accuracy) in classifying the potato leaf health status, however, exhibited the longest time of processing. Optimized K-nearest neighbors (KNN) have a considerable accuracy (93.21%) and inference time (32.63 s). For PDA prediction, hybrid decision tree and support vector machine regression (HDT-RSVM) defeated other feature-based machine learning models. Ultimately, the formulated DT:SVM:RSVM model offers accurate disease identification and quantification on the potato leaf surface by using a consumer-grade camera that is translatable to low-income agricultural units.