IGWO-Driven DeepForestNet: A Hybrid Shallow Fusion Model for Multiclass Leaf Disease Classification

Abstract

Modern deep learning (DL) techniques, especially convolutional neural networks (CNNs), have greatly increased the precision of multiclass classification tasks used to diagnose plant diseases. To detect multiclass leaf diseases in corn and apple plants, this study aims to develop a reliable and effective model. The proposed DeepForestNet integrates mid-level feature representations from DenseNet201 and MobileNetV2 through a shallow fusion mechanism. The fused feature vector is classified using a Random Forest (RF) classifier to enhance generalization and robustness. To further optimize model performance, the hyperparameters are fine-tuned using the Particle Swarm Optimization (PSO) and Improved Grey Wolf Optimization (IGWO) algorithms. The effectiveness of these optimization techniques is validated through comparative statistical analysis using the Wilcoxon signed-rank test and convergence evaluation. The IGWO-optimized DeepForestNet achieves a classification accuracy of 97.59% for apple leaf diseases (13 classes) and 97.19% for corn leaf diseases (4 classes), outperforming baseline and PSO-tuned variants. The proposed IGWO-driven DeepForestNet demonstrates high accuracy, stability, and convergence, making it a promising framework for automated and precise plant leaf disease identification in precision agriculture.