A NOVEL DEEP LEARNING FRAMEWORK FOR PLANT DISEASE DETECTION USING EFFICIENTNET MODEL

Abstract

Global agricultural output is seriously threatened by plant diseases, which must be identified quickly and accurately to minimize crop loss and improve farming methods. Outdated methods of disease identification, relying on expert knowledge and manual inspection, are time-consuming and prone to errors, making it challenging to address large-scale agricultural monitoring. To address these challenges, this work presents a deep learning (DL)-based system for accurate plant disease detection using the EfficientNet architecture. The system begins with data collection, where a plant disease dataset of healthy and diseased plant leaves is gathered. Data preprocessing follows, including image resizing for uniformity and data augmentation to increase the dataset size and diversity. Once the data is pre-processed, feature extraction is performed using Histogram of Oriented Gradients to capture key texture and edge features from the images. The EfficientNet model is then selected and trained on the extracted features for plant disease classification. The results show that the training accuracy reached 0.988, and the training loss dropped to 0.033, with similar performance observed for validation, where accuracy reached 0.988 and loss also converged to 0.033, indicating strong model performance. This work proposes a computationally efficient, accurate DL-based framework for plant disease classification, demonstrating both high accuracy and generalization capability for applications in agriculture.