An implementation and analysis of deep learning models for the detection of wheat diseases

Abstract

Detecting a disease visually is a time-consuming and error-prone operation, and in the agricultural arena, for disease control, crop yield loss prediction, and global food security, automatic and accurate evaluation of disease severity in crops is a particularly demanding study area. Deep Learning (DL), the latest innovation in the era of Artificial Intelligence (AI), is promising for fine-grained categorization of crop diseases since it eliminates labor-intensive feature extraction and segmentation. To diagnose the disease from photos, multiple pretrained models which are ResNet50, EfficientNetB0 and InceptionV3 along with ViT and a hybrid CNN model have been trained on the wheat disease dataset. Again, an ensemble model of the hybrid CNN and the ViT has been proposed which has been compared with all the other models and the proposed model gets the highest accuracy of 99.34% among all the models.