A stacked deep learning approach for multiclass classification of plant diseases

Plant diseases are one of the main factors affecting food production and reducing production losses, they must be swiftly identified and treated. Deep learning algorithms in association with computer vision techniques have recently found usage in the diagnosis of plant diseases, offering a potent tool with highly accurate results. The objective of this study is to identify a stacking ensemble-based solution by using several algorithms in the process of classifying and diagnosing plant diseases, describing trends, and emphasizing gaps.

The stacking ensemble is made using top four performing deep learning algorithms and multi-layered perceptron as meta classifier. In this regard, we reviewed more than 15 studies from the previous three years that address problems with disease detection, dataset characteristics, researched crops, and pathogens in various ways.

The proposed ensemble model achieved a maximum accuracy of 98.13% compared to the conventional architectures. For comparing the results, various performance metrics are used such as accuracy, loss, precision etc. which outperformed the results of the deep learning algorithms run separately for the data as shown in Table 5.

The suggested framework can help identify the presence of disease in a sample of plant leaves as a preventative strategy as the results were quite promising compared to the results of existing literature.