Advanced Deep Learning Approaches for Image-based Diagnosis of Banana Leaf Diseases

Abstract

Early detection of plant leaf diseases is beneficial as it helps agriculturists to apply remedial measures well in advance. This is highly recommended for a good yield from crops, which enhances the economy of an agriculture-based country. Computer vision and deep learning techniques, used in the field of precision agriculture, facilitate early detection and classification of plant leaf diseases. The literature proclaims that deep learning models outperform machine learning approaches for the classification of leaf diseases. In this paper, the state-of-the-art deep learning methods for detection and classification are applied on banana leaf dataset. The AlexNet, VGG19, DenseNet201, ResNet50, and MobileNetV2 convolutional neural networks are the models compared in this paper. The real-time images of banana leaves are used to train and test these models using Python programming. Healthy and two common diseases of banana leaves, namely Leafspot and Sigatoka, are classified in this work. After data augmentation and preprocessing, all the models could achieve good testing accuracies of more than 90.6% in the 80 of training, 10% of validation, and 10% of testing sets. ResNet50 deep technique outperforms the other architectures in 80% of the training set. The training and testing accuracies depend on the data augmentation and image pre-processing techniques.