A Comparative Survey of Maize Leaf Diseases using Pre-Trained Convolutional Neural Networks

Abstract

Transfer learning technique reduces the time for design and development of classification models. In this research, we developed maize leaf disease classification models using the most recent and widely used transfer learning techniques. Six pre-trained models were used for developing the maize leaf disease classification models, such as VGG19Net, ResNet152, InceptionV3Net, MobileNetV2, DenseNet201, and NASNetLarge. The pre-trained model was trained on the maize disease classes of the standard plant leaf disease dataset for 100 training epochs. There are 200 images that were used to test the performance of the classification models on maize leaf disease detection. Standard performance metrics such as classification accuracy, precision, recall, and F1 score were used to estimate the performance of the models. Densenet201 achieved a classification accuracy and an average F1 score of 99.25% and 98.5%, respectively. The performance metric scores of Densenet201 on maize leaf disease detection are better than those of other classification models. The research identified the classification performance of the Densenet201 as superior to VGG19Net, ResNet152, InceptionV3Net, MobileNetV2 and NASNetLarge on maize leaf disease classification.