Fully Automatic Wheat Disease Detection System by Using Different CNN Models

Abstract

AbstractIn India, agriculture assumes a significant part in view of the rising number of individuals and extended interest in food. In the world, Wheat is the third most gathered and consumed grain. One critical effect on wheat crop yield is a disease because by parasites, contaminations, and microorganisms. So that is the justification behind a huge piece of the wheat crop becoming spoiled. Multiple dozen wheat contaminations are risky to the yields, so it is very important to detect these diseases at the exact time. The determination of the diseases is done with a visual investigation by specialists and an organic assessment is a subsequent option if fundamental. It is an extremely tedious and costly procedure. Deep learning is the only way by which one can solve such problems, this branch additionally considers the early recognition of wheat diseases by applying convolutional neural networks (CNNs) close to the well-known models. In the current work, we consider four different classes of wheat pictures which contain tan spot, fusarium head blight, stem rust, and healthy wheat. We execute the different CNN models to the gathered dataset. We use number of parameters to train these models are: loss function = “categorical cross-entropy”, activation function = “softmax”, optimizer = “adam”, batch size = 64, epochs = 80. After applying these models to the dataset we examine that the ResNet model delivered the best calculable outcomes. The proposed approach has gotten the most noteworthy classification exactness of 98% utilizing the ResNet50 model when contrasted with MobileNet, and DenseNet models. This shows that deep learning has shown generally excellent execution for the order of various illnesses. We did this execution with the assistance of google colab.KeywordsDeep learningWheat diseaseConvolutional neural network (CNN)Pathogens