VGG16-3PGA: A Hybrid Approach for Plant Disease Detection

Choosing significant features of on-field plant images and effectively classifying diseases is a challenging endeavor. This is owing to the complexities of on-field plant images, which can be influenced by a wide range of variables such as illumination, soil conditions, plant growth phases, and environmental influences. Feature selection is regarded as an optimization problem employing an efficient optimizer to choose an informative feature subset from a more extensive set of features. The noisy, redundant, and irrelevant features of the dataset impede the image classifier’s effectiveness. Handcrafted feature extraction techniques produce high-dimensional features. As a result, feature extraction and feature selection approaches must be explored for enhancing image classification performance. With feature extraction approaches, the raw image data is converted into a set of useful features that can be employed for classification. In this paper, we introduce a novel hybrid approach for image classification that leverages deep transfer learning with VGG16 for feature extraction and the three-parent genetic algorithm (3PGA) for feature selection. The most pertinent and informative features for image classification are obtained by combining feature extraction and feature selection strategies. Our approach is designed to improve accuracy and reduce training time by selecting only the most important features. We tested the proposed plant disease detection approach on rice, cassava, tomato, potato, and pepper bell crops. The experimental results demonstrate that our approach outperforms other popular methods, including the Genetic Algorithm and VGG16-Support Vector Machine (SVM) etc, in terms of accuracy, precision, recall, and F1 score.