Optimized Fractal Fusion Driven Deep Neural Network: An Efficient Hybrid Methodology for Disease Detection and Classification in Plant Leaves

Abstract

Plant diseases and pest attacks are among the major issues in agriculture, which have a great impact on crop growth and yield. Early and correct detection of these diseases is vital for effective management and keeping plant health at the desired level. This research proposes a new method for detection and classification of plant diseases, the Optimized Fractional Fusion Deep Neural Network (OFF_DNN) model. The procedure uses a synthetic dataset created by Generative Adversarial Networks (GANs) to enhance the classification results. The OFF_DNN model first enhances low-resolution plant images to produce high-quality images. After this, the enhanced images are GAN-synthesized to create additional data for effective training. After this, two segmentation methods, i.e., Marker-Controlled Region-based Thresholding (MCT) and Active Lesion-based Multilevel Segmentation (ALMS), were performed to segment the regions of interest in the enriched images. Fractal fusionbased feature extraction on these regions was done, and Particle Swarm Optimization was applied for the optimization of the features. The optimized features were then fed into the model as input features for the DNN model for the classification of diseases. The experimentation shows that the OFF_DNN model is effective, achieving average accuracy rates of 98.64, 98.79, 98.6, and 0.9885%, providing an opportunity for utilizing this in automated plant disease detection in agriculture. Hence, incorporation of image enhancement, synthetic data creation, and optimization of features addresses salient challenges in plant disease management, providing a feasible solution to improve yield and quality.