Performance Analysis of Pre-Trained Deep Learning Architectures for Classification of Corn Leaf Diseases

The agriculture sector plays a crucial role in supplying high-quality food and contributes the most to expanding economies and populations. Plant infections can cause significant harvest losses and the extinction of species. The swift identification of plant illnesses by means of precise or automated detection techniques can improve the quality of food production and reduce economic losses. In recent years, deep learning has significantly enhanced the accuracy of image classification and object detection systems' recognition capabilities. This study discovers four distinct leaf types on the leaves of corn and classifies them according to the plant diseases that cause them. These leaf varieties are referred to as blight, common rust, grey leaf spot and healthy leaves. The number of samples collected for each is 1306, 574, 1146, and 1164, respectively. By extracting deep features, three distinct enhanced deep neural networks (DNNs) are used to classify corn leaf disease. The validation accuracy, test accuracy, prediction, recall, F1 score, and area under the curve (AUC) are analysed to determine performance. From the extensive investigations, it is noted that the classification accuracy for the models is as follows: 92.6% for CNN, 94.3% for AlexNet, 95.7
^%
for SqueezeNet, and 97.3% for Xception. The findings indicate that the models produced during the investigation phase to identify the leaf disease were superior to those of any current machine learning (ML) classifiers.