Feature-Augmented Convolutional Neural Network Optimisation for Wheat Yellow Rust Identification

Abstract

Nowadays, yellow rust is a type of condition having a huge domain triggering excessive adverse effects on wheat. Manually addressing wheat yellow rust using the standard approach is not highly effective. In order to enhance this condition, a Deep Learning (DL)-based method was applied in this analysis to identify wheat yellow rust from the Yellow Rust-19 dataset leaf images. The solution offered was constructed using the Optimized Feature-Augmented Convolutional Neural Network (OFAC-Net) model to classify R (Minor infection), MR (Small and Medium Infection), MR-MS (Moderate Resistance and Susceptible), MS (Medium Infection), and S (Major signs of infection), aiming to improve the efficiency of detecting and classifying yellow rust in wheat leaf images. The implemented OFAC-Net models integrate Principal Component Analysis (PCA) used for feature extraction and reduce the dimensionality, alongside hybrid feature selection techniques utilizing the Whale Optimization Algorithm (WOA) and Firefly Algorithm (FFA) to refine the extracted features and boost accuracy. This combination addresses existing challenges and improves classification performance, with the results being classified through a Convolutional Neural Network (CNN). The implemented model attained a classification Accuracy of 98%, Precision of 98.76%, Recall of 99%, and Mean Square Error (MSE) of 1.98%. The dataset, curated from the several severity levels of yellow rust disease, in wheat, consists of a total of 15,000 wheat leaf images. It was designed to support the classification tasks, with leaf images evenly divided across severity levels of yellow rust disease in wheat. These results highlight the techniques` superiority over traditional machine learning (ML) techniques and other advanced models. The proposed OFAC-Net model is a promising solution for real-time agricultural applications, offering both high performance and computational efficiency suitable for mobile and edge devices.