Hybrid CNN Feature Fusion with Optimization for Precision Potato Leaf Disease Classification

• Umar A. IBRAHIM

  Department of Computer Science, Sule Lamido University, Kafin Hausa, Jigawa, Jigawa State, Nigeria

  Author

• Abdulra’uf G. SHARIFAI

  Department of Computer Science, Northwest University, Kano, Kano State, Nigeria

  Author

Potato production is highly vulnerable to a range of diseases that threaten global food security and agricultural productivity, particularly in uncontrolled farming environments. This study developed a hybrid deep learning framework for potato leaf disease classification, integrating multi-model deep feature fusion from five pre-trained convolutional neural network (CNN) backbones (VGG19, ResNet50, DenseNet121, InceptionV3, and MobileNetV2) with a two-stage hybrid resampling strategy. (Borderline-SMOTE and SMOTETomek) to address severe class imbalance. Feature selection was performed using a Modified Walrus Optimization Algorithm (mWAOA) enhanced with genetic operators, followed by Principal Component Analysis (PCA) to retain 95% variance while reducing computational complexity. The optimized feature set was classified using a fully connected neural network. Experimental results demonstrated a recall of 99.68%, an accuracy of 98.68%, and consistently high precision, and F1-score values, surpassing individual CNN baselines and prior published models. The proposed framework significantly improved minority class detection and robustness under varying environmental conditions. These findings highlight its potential for scalable, real-time disease monitoring and precision agriculture applications.

• FJET_12_91_84

Copyright (c) 2025 FUDMA Journal of Engineering and Technology

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.