Identification of Cherry Powdery Mildew Using Deep Convolutional Neural Network

Agriculture is essential for humanity’s survival, and productivity is crucial in agriculture. Owing to its multiple nutrients, cherry has become an important fruit for daily consumption; however, crop productivity is mainly reduced by powdery mildew. Hence, recognizing this disease is vital to farmers. We developed a deep convolutional neural network (DCNN)-based method to assess cherry tree health using leaf images. We constructed the DCNN model on Keras with TensorFlow, preprocessing the original images for convenience. We used a visualization method to analyze intermediate model layers, comparing features of healthy and diseased leaves. The model achieved 99.2% accuracy after 10 training epochs. Misclassification occurred when leaf shadow edges were close to actual leaf edges, leading the model to mistake actual edges as features indicating disease, offering new insights for distinguishing the disease. Simulation results demonstrate that identifying plant disease via a DCNN-based protocol is suitable and adaptable to other plants, providing high practical value.