EfficientNetB0-Based Deep Learning Framework for Automated Eggplant Leaf Disease Classification

Abstract

 Early and accurate detection of plant diseases is crucial for enhancing crop yield and sustainable agricultural practices. This work presents an automated classification of eggplant leaf diseases using the EfficientNetB0 deep learning architecture. The model is built to detect four different classes,  namely Healthy Leaf, Insect Pest Disease, Leaf Spot Disease, and Mosaic Virus Disease. A publicly available dataset was used, and pre-processing all images in terms of resizing and normalization was done to match the input requirements of EfficientNetB0. Transfer learning using ImageNet pre-trained weights was used to improve feature extraction and decrease the training complexity. The dataset was split into train, validation, and test data using a stratified split strategy. Experimental results show that the accuracy of the overall classification is 88.5% with balanced precision, recall, and F1-scores in all classes. The confusion matrix analysis proves reliable class-wise performance with minimal misclassification. The results show that EfficientNetB0 is an effective and computationally efficient method for eggplant leaf disease detection, and has great potential to be used in real-world agriculture.