Detection of Powdery Mildew in Cherry Plant Leaves using Machine Learning and CNN

Timed observation and accurate detection of plant diseases is vital for minimizing the damage to the crops and securing food for the long term, but particularly with cherry powdery mildew deeply affecting cherry production. The paper gives a comparison of traditional machine learning techniques, including Logistic Regression, Naive Bayes, Decision Tree, and Support Vector Machine, with a custom Convolutional Neural Network with classification of cherry leaf between healthy and powdery mildew affected. Using a preprocessed and augmented image dataset split into training and testing sets, models were evaluated via accuracy, F1-score, specificity, sensitivity, ROC AUC, and confusion matrices. Support Vector Machine achieved the best result with the highest accuracy of 96.11% using color histogram features. Convolutional Neural Network shows strong discriminative ability but has room for improvement. This study shows the strong key points of precision agriculture, with chances of refinement of the Convolutional Neural Network to enhance plant disease detection and enhance agriculture.