DELM: deep ensemble learning model for multiclass classification of superresolution leaf disease images

Abstract

Tomato plant leaf diseases are the main risk factor for plant growth. Detection of diseases at the initial stage is the main and most complex task for farmers due to common morphological properties like colour, shape, texture, and edges. The quality and quantity of agricultural products might be significantly reduced because of plant diseases. One hundred and seventy-seven pathogens, including 167 fungi and others like bacteria, algae, and nematodes, attack the well-known tomato plant. Postharvest illnesses may also result in significant productivity losses. An expert opinion is needed for illness analysis due to the modest variations in the symptoms of different tomato diseases. Farmers who misuse pesticides may suffer financial losses as a result of erroneous diagnoses. Plant leaf diseases are difficult to categorize since there are many similarities among different groups and intricate design changes. In this paper, the authors present a deep ensemble learning model (DELM) for autonomous plant disease identification. The pretrained models are refined using transfer learning. Different augmentation techniques, including picture enhancement, rotation, and scaling, combat overfitting. This research gives a thorough taxonomy of the performance of a single model and various ensemble learning models to classify superresolution tomato plant leaf disease images. With the publicly available dataset, including ten different biotic disease classes of tomato plants, the efficiency of the projected prototype is evaluated. The efficiency of a single pretrained model, VGG16, shows 98% accuracy with an F1-score of 93.25%. In addition, an ensemble of three models (VGG16, InceptionV3, and GoogleNet) shows more accurate results than other ensemble models.