Maize disease identification based on optimized support vector machine using deep feature of DenseNet201

Abstract

In recent times, maize diseases have become widespread globally, adversely impacting agricultural productivity and causing significant financial losses. Recognizing these diseases in maize leaves presents formidable challenges due to the complexities involved in extracting features from a dynamic environment. Issues such as inconsistent lighting conditions, reflections from light sources, and other variables add to the intricacies of this task. This paper introduces a novel classification model that leverages deep features and advanced optimization strategies to categorize four variations of maize leaves: blight, common rust, gray leaf spot, and healthy. Our approach harnesses the power of DenseNet201, a specialized deep-learning architecture tailored for image-classification tasks. This model excels in extracting crucial features from maize leaf images. Furthermore, we meticulously fine-tuned the Support Vector Machine (SVM) using Bayesian optimization techniques to discern and categorize diseases based on these extracted features. In our experimental assessment, we meticulously curated a dataset comprising 4988 images across four distinct maize leaf classes. The outcomes of our experiments are highly compelling, showcasing that our proposed model achieved an outstanding classification accuracy of 94.6%. Notably, this accuracy surpasses the performance attained by an SVM operating without the assistance of deep features and optimization techniques.

Keywords

• Maize leaf diseases
• Deep features
• DenseNet201
• SVM
• Bayesian optimization
• Hyperparameter tuning