Leveraging Autoencoders for Accurate Plant Disease Diagnosis in the Face of Unbalanced Data

Deep learning has emerged as a powerful tool in addressing real-world challenges in the field of agriculture, and one of the most pressing issues faced is the imbalance of data, which complicates data modeling. In this study, we introduce an innovative hybrid model based on autoencoders to tackle this problem head-on. Our research focuses on the classification of images depicting six distinct categories of brinjal crops: Early Blight, Little Leaf, Phomopsis Blight Leaf, Phomopsis Blight Fruit, Healthy Leaf, and Healthy Fruit. To accomplish this, our hybrid model harnesses the strengths of six specialized autoencoders, each meticulously designed to handle a specific class. These autoencoders are responsible for extracting encoded representations of the input data. These encoded representations are then merged and processed through a series of convolutional layers, followed by flattening, dropout, and a dense layer. This comprehensive approach to feature extraction, made possible by the autoencoders, leads to an impressive classification accuracy rate of 91%. By effectively addressing the challenge of imbalanced data in agriculture, our novel hybrid network offers a promising solution to improve the accuracy and efficiency of classification tasks in this vital field.