Identifying corn leaves diseases by extensive use of transfer learning: a comparative study

Abstract

Deep learning is currently playing an important role in image analysis and classification. Diseases in maize diminish productivity, which is a major cause of economic damages in the agricultural business throughout the world. Researchers have previously utilized hand-crafted characteristics to classify images and identify leaf illnesses in Maize plants. With the advancement of deep learning, researchers can now significantly enhance the accuracy of object classification and identification. Using the "Corn or Maize Leaf Disease Dataset" from the Kaggle website, four forms of maize leaf diseases were investigated: blight, common rust, gray leaf spot, and healthy. The pictures obtained from these corn leaf illnesses are categorized using four deep convolutional neural network (CNN) models that have been pre-trained (GoogleNet, AlexNet, ResNet50 and VGG16). Accuracy, precision, specificity, recall, F-score, and time are the six metrics used to assess the performance of any transfer learning (TL) model. MATLAB programming software is used to design and train the TL models. The accuracy of each item in the dataset has been checked. It has been determined that GoogleNet, AlexNet, VGG16, and ResNet50 each have an accuracy of 98.57%, 98.81%, 99.05%, and 99.36%, respectively.