EfficientNetB3-Adaptive Augmented Deep Learning (AADL) for Multi-Class Plant Disease Classification

Abstract

Plant diseases can significantly impact agricultural productivity if not promptly identified and treated. Traditional plant disease classification methods are often challenging and time-consuming, making the identification of diseases a challenging task. This paper aims to bridge research gaps and address challenges in existing methodologies by proposing an efficient, effective multi-class plant disease classification approach. The research explores the application of pre-trained deep convolutional neural networks (CNNs) in this classification task, utilizing an open dataset comprising 52 categories of various diseases and healthy plant leaves. This study evaluated the performance of pre-trained deep CNN models, including Xception, InceptionResNetV2, InceptionV3, and ResNet50, paired with EfficientNetB3-adaptive augmented deep learning (AADL) for precise disease identification. Performance assessment was conducted using parameters such as batch size, dropout, and epoch counts, determining their accuracy, precision, recall, and F1 score. The EfficientNetB3-AADL model outperformed the other models and conventional feature-based methods, achieving a remarkable accuracy of 98.71%. This investigation highlights the potential of the EfficientNetB3-AADL model in offering accurate, real-time disease diagnostics in agricultural systems. The findings suggest that transfer learning and augmented deep learning techniques enhance the accuracy and performance of the model.