Novel Plant-Pathogen Classification Techniques based on Supervised Machine Learning Techniques using Optimized Gabor Sets

Summary: —Plant diseases are a key cause of our food insecurity, which is expected to increase because of globalization, global warming, and other factors. Plant disease classification is thus an important step in the control of plant diseases. Traditionally, direct, and indirect procedures such as RT-PCR or thermography were used, which are costly, need a qualified expert, and do not provide a real-time diagnostic. Plant diseases are caused by a variety of pathogens and recognizing them using machine learning is a significant focus of this research because there hasn’t been much work done in this area of plant health monitoring, and early detection of the pathogen can provide insights into how to best treat the infestation. We employed computer vision techniques to extract features of diseased leaves from the PlantVillage dataset and produced promising results utilizing several Supervised Machine Learning techniques such as Random Forest, SVM, Naive Bayes, and KNN to classify the pathogen type. We also focused on optimizing the dataset by finding the best-suited filters for identifying plant pathogens using a novel technique we developed to organize the filters in the best possible order, as well as determining the best hyperparameters for our model.