Deep Convolutional Neural Network based Wheat Leaf Disease Detection in Precision Agriculture

Wheat is one of the fundamental crops that sustains billions of people globally. Unfortunately its production faces major challenges from numerous plant based diseases which can drastically decrease yields as well as the economic returns. There exist a variety of frameworks utilising computer vision and deep learning techniques which can be invaluable for numerous tasks aiding the stakeholders in agriculture and crop management by identifying diseased sections of plants.Traditional methods to detect the disease are prone to human errors and such limitations underscore the need for automated and reliable disease detection systems. Also AI is likely to predict a future outbreak of any such disease by analysing the historical and geographical data. In this paper, a novel technique have been proposed whose novelty lies in the result that has been achieved using the proposed algorithm which when compared to two other deep learning models that are ResNet50 and VGG16 achieved a higher accuracy. The deep learning approach for wheat leaf disease detection implemented in this paper is performed using a convolutional neural network (CNN) architecture implemented in Python, OpenCV, NumPy, and TensorFlow alongside implementing some data augmentation leveraging the power of deep learning to extract complex features from wheat leaf images enabling accurate classification between healthy and diseased leaves. The aforementioned algorithm has been trained using a default Crop Disease Dataset extracted from a Github repository which has been able to achieve an accuracy of $88.53 \%$ with one full cycle of 30 epochs recording its final epoch with $91.04 \%$ accuracy and the final K Fold procedure resulting in $91.42 \%$ accuracy.