Classification of Plant Leaf Disease Using Ensemble Deep Transfer Learning

Abstract

Crops can be affected by various types of pathogens that cause diseases, leading to significant damage and negative impacts on food production and quality, causing farmers financial losses. Therefore, it is essential to detect and prevent plant diseases at their initial stages promptly. Unfortunately, the process of detecting and controlling plant diseases can be challenging for farmers. However, deep learning techniques can potentially make a significant contribution by accurately classifying plant diseases at their earliest stages. To overcome the limitations of previous research, this work proposes a new method for diagnosing and classifying plant leaf diseases. The proposed approach enhances the classification of popular crops such as tomato, potato, and pepper by utilizing a dataset that comprises nine categories-three of which are healthy and the other six of which are infected-using transfer ensemble learning with MobilenetV3 small and Resnet50. The model has only 37 million training parameters, which saves time and computational power and reduces overfitting while achieving an accuracy of 99.50%. The model's ability to address false negatives and false positives ensures reliable and accurate plant disease classification at an early stage. The proposed method can work with any image dataset and achieve high performance, and it has been validated on the benchmark open dataset Plantvillage, demonstrating excellent performance for the proposed diagnosis approach.