Classification of Maize leaf disease using improved deep convolutional neural network

The classification of diseases affecting maize is a prominent area of research because it affects agricultural production significantly. While the classification of maize diseases has improved considerably due to deep learning (DL) approaches, problems including low precision and detection challenges persist. Farmers must identify maize leaf (MLf) diseases with comparable symptoms as soon as possible and with clarity. To classify MLf disease, this study proposed an improved transformer network called Vanilla transformer (VaT). We modified the top layer of the VaT model. By applying techniques for image preprocessing and augmentation, the disease-affected area has been detected. Following that, EfficientNetB7 and MobileNetV3 were used to extract features from images. Then, the performance metrics are calculated to assess the classifiers’ performance. By comparing findings with recently published relevant research, the improved VaT obtains 99.96% classification accuracy and the Efficient-NetB7 achieves 99% rate of classification, which appears to be satisfactory and promising.