Fusion of AI Methods for Enhanced Plant Leaf Disease Classification

Abstract

Plant leaf disease classification is essential to save the global food supply and boost agricultural productivity. Some new advancements in this field have improved this diagnosis with efficiency and accuracy. In this work, a novel hybrid method is proposed by fusing conventional machine learning algorithms, such as Random Forest (RF) and Support Vector Machine (SVM), with advanced techniques like Convolutional Neural Networks (CNN). The two hybrid versions are achieved with the help of CNN+RF and CNN+SVM. Particle Swarm Optimization serves as the feature selection method for achieving better classification accuracy. PSO extracts the most valuable features coming from each of the classification models and enhances the ability to distinguish between healthy and diseased plants by optimizing the feature subset for each model. It's a very vast dataset of plant leaf images showing a diseased leaf. The experimental results of the targeted hybrid approach are compared with that of single classification methods. The hybrid models present the benefits of synergy in the form of high rates of accuracy with improved generalization. On the other hand, the PSO feature selection process considerably enhances the classification results of the classifier by revealing the discriminative potential of selected features. This is an entirely novel hybridized strategy that presents and elevates much potential toward plant leaf disease classification. The results underline potential strategies for efficient plant disease management toward enhancing agricultural productivity and fostering sustainable farming. This new hybrid architecture proposed considering this existing issue, can be viewed as a benchmark for other domains in which such classification tasks can be performed, which has an overall advanced view of the benefits of applying two methodologies together to enhance performance and accuracy. When compared against CNN+RF, the experimentation of CNN+SVM has shown a 92% F1 score, 92% recall, 93% accuracy, and 91% precision and recall.