Plant Disease Classification Using Multi‐Feature Extraction and a Hybrid Learning Rate Scheduler in Deep Learning Models

ABSTRACT

Accurate plant disease detection is essential for improving agricultural productivity and ensuring food security. This study proposes a hybrid classification framework that integrates handcrafted features (GLCM, LBP and colour) with deep features extracted using EfficientNetB2. To reduce dimensionality and enhance efficiency, Principal Component Analysis (PCA) is applied. A custom dense neural network with a fuzzy logic–based output layer is introduced to handle uncertainty in disease patterns. Additionally, a hybrid learning rate scheduler is proposed to improve training stability. The model is evaluated on multiple datasets, including real‐world images from Basra, Iraq, achieving high accuracy and demonstrating strong robustness under diverse environmental conditions.