Hybrid Machine Learning and Deep Learning CNN Models for Apple Trees Leaves Pathologies Classification

Agricultural development in recent years has been based entirely on smart management. Leaves diseases on apple trees is a very common problem. The present work is a continuation of previous research of the author's team to solve it. The recently proposed apple leaves pathology prediction model based on deep learning convolutional neural network (CNN) is extended to hybrid models. The latter are a combination of a deep learning CNN and classical machine learning techniques. To increase the classification accuracy two hybrid models are proposed: CNN+SVM (Support Vector Machines) and CNN+RF (Random Forest). The CNN is based on modified DenseNet architecture and SVM and RF models are integrated as CNN output is fed to SVM and RF models respectively to ensure higher classification accuracy. The suggested hybrid models are evaluated on a data set that comprises 1800 images of pathologies in apple trees’ leaves. The experimental results prove the higher accuracies of the hybrid deep + machine learning models compared to the deep learning CNN model.