Performance of Darknet CNN Architecture in Wheat Disease Classification

This study focuses the application of the Darknet-53 convolutional neural network (CNN) architecture for the classification of wheat diseases, specifically yellow rust, healthy wheat, and brown rust. Utilizing a dataset comprising 924 images of yellow rust, 1116 images of healthy wheat, and 902 images of brown rust sourced from Kaggle, the data were split into training and testing sets following a traditional 80/20 split during pre-processing. The model was trained using Stochastic Gradient Descent with Momentum (SGDM) as the solver, an initial learning rate of 0.01, a batch size of 64, and a validation frequency of 50, leveraging GPU capabilities. The resulting model demonstrated a classification accuracy of 95%, validated through confusion matrix analysis and key performance metrics, including precision, recall, and F1 scores. High precision, recall, and F1 scores were achieved for Brown Rust and Healthy Wheat classification, illustrating the model’s robustness. The slightly lower recall for Yellow Rust indicates an area for potential improvement.