AppleCNN: A new CNN-based deep learning model for classification of apple leaf diseases

Abstract

Day by day the world's population is increasing and the land people use for food is decreasing. Fruit trees in existing agricultural lands are under constant threat from numerous pathogens and insects. Therefore, continuous monitoring is important to ensure maximum yield. Apple is a very important fruit both in terms of consumer demand and global trade. However, apple growth, quality and yield can be affected by a number of diseases. The key to successful disease management and prevention of further outbreaks in apples is early and accurate identification of the disease. If apple foliar disease is not identified early, it can lead to overuse or underuse of chemicals. This can lead to increased production costs and adverse effects on the environment and health. Apple leaf diseases are grouped into 4 different classes: apple scab, cedar apple rust, healthy apple and complex disease symptoms (more than one disease on the leaf). A new CNN model is proposed by using pre-trained VGG19, DenseNet169, MobileNetV2, Xception and NASNetLarge architectures as input layer. This proposed CNN model consists of 23 layers based on computer vision preprocessing techniques and deep learning. With the proposed CNN model, 98% success rate is achieved for apple fruit disease class.