DenseNet-ResNet-Hybrid: A Novel Hybrid Deep Learning Architecture for Accurate Apple Leaf Disease Detection

The accurate identification of diseases on apple production is an important issue due to the worldwide importance of apple production in contemporary agriculture. Identifying diseases correctly can be challenging and affects food safety and economic loss significantly. To alleviate this, deep learning approaches, and particularly Convolutional Neural Networks (CNN), have been able to provide new and reasonable options in the agricultural field. In this study, there is a hybrid model proposed, called DenseNet-ResNet-Hybrid, which brings together architectures from DenseNet and ResNet, to provide an improvement in the extraction of features together. It has been designed to fuse the inherent capabilities of DenseNet and ResNet, capturing both detail features and deeper level features in apple images, to enhance the ability to separate diseases that are overlapped with the producer's natural environment (e.g. overlapping leaves/fruits). We finally show two complete comparative experiments against two popular models (like VGG16, ResNet50, Inception-v3) under the exact same conditions to demonstrate the strength of their ability to accurately classify apple leaf diseases with consistency. We use a broader select of image types to demonstrate our work, and ultimately suggest our proposed hybrid model demonstrates competitive performance in accurate classification on apple images on the whole.