Fine-tuned Convolutional Neural Network Model for Wheat Leaf Disease Classification

The pathogenic fungus can cause devastating damage to cereal crops and may result in a wide range of cultural influences. The large-scale eradication of the diseased crop is not a solution and is also challenging due to the problem of economic loss and limited manpower. Disease detection based on the assessment of crop images can be an effective way of reducing and stopping the spread of the disease in the entire field. In the proposed study, a convolutional neural network model for the early detection of five distinct fungal diseases (leaf rust, stem rust, yellow rust, powdery fungus, and septoria) in wheat crops has been done that may result in reducing the spread among other plants. The model has been implemented with a dataset of 1972 images of wheat fungus diseases collected from Kaggle. The model has been trained and tested at different number of epoch values and the results have depicted that the proposed model shows the highest accuracy of 98.83% at epoch 12. The loss value has resulted that during the training and testing, in the initial epochs till 6, there was a huge difference in the loss, when, the epoch value started to increase the loss of both training and testing started to decrease. These results have been further compared with other models which resulted in the proposed model having the highest accuracy in comparison to existing models.