Orchard Guard: Deep Learning powered apple leaf disease detection with MobileNetV2 model

Abstract

The apple crops are susceptible to various diseases that can substantially reduce quality and yield, emphasizing the need for an accurate and automated detection system. The designed model can efficiently detect four different classes of an apple leaf viz. Apple Scab, Black Rot, Cedar rust, and Healthy. The detection has been carried out using a transfer learning approach with different models such as AlexNet, DenseNet121, ResNet-50, and MobileNetV2 as the primary ones. With hyperparameter tuning and by using different optimizer combinations we trained the MobileNetV2 model to achieve the best accuracy. The selected model is trained and fine-tuned on an Apple dataset of 3175 images, leveraging transfer learning from pre-trained models on large-scale image datasets. The designed ‘Orchard Guard’ model has achieved an accuracy of 99.36%. The research findings can help in the selection of a useful model for actual use in orchards and can aid in the creation of effective and precise disease management systems.