A comprehensive survey on cassava disease detection and classification using deep learning models

In the past few years, methods using deep learning have demonstrated promising outcomes in a variety of image-based applications, including the identification and classification of plant diseases. In this article, we propose a comparative examination of deep learning models for the identification and categorization of diseases affecting cassava leaves. The dataset used in this study comprises non balanced samples, posing a challenge due to imbalanced class distribution. Our research focuses on investigating the performance of different deep learning architectures, including Transformer Embedded ResNet, EfficientNetV2, with visual attention mechanism, and a mobile-based deep learning model, in addressing this problem. The suggested models use deep convolutional neural networks (CNNs) to their full potential and incorporate a variety of deep learning techniques, such as transformers and attention mechanisms, that improve accuracy as well as efficiency. Through extensive experiments, we analyse each model's performance in terms of classification accuracy, precision, recall, and F1-score. Moreover, we compare the computational complexity and deployment feasibility of these models in real-world scenarios. Conclusions demonstrate the effectiveness of the suggested models accomplish significant improvements in cassava disease detection and classification compared to traditional techniques for machine learning. The deep learning models effectively handle the non-balanced dataset and exhibit robustness in identifying different types of cassava leaves disease. Our survey provides Informative data about the suitability and effectiveness of deep learning techniques for accurate and efficient plant disease diagnosis