Abstract

Plant diseases have a major effect on the productivity and food security of the global community. The latest achievements in the field of deep learning allowed the creation of the automated system of plant disease detection based on convolutional neural networks (CNNs). Nevertheless, whether these models can be interpreted is an issue. This paper has used pre-trained models based on convolutional neural network (CNN) to identify plant diseases efficiently. We paid attention to optimization of the hyperparameters of sophisticated pre-trained models, including ConvNeXt, EfficientNet-B7, SE-ResNet, and SE-DenseNet. The experiments were conducted on the popular PlantVillage dataset (54,305 image samples of the various plant disease species). The models results were measured by the classification accuracy and sensitivity, specificity and F1 score. The comparison was also conducted with the similar state-of-the-art studies. The experiments proved that the SE-DenseNet had 99.81% classification accuracy that was better than other state-of-the-art models.

Authors

Sarita Singh, Noopur Goel
Veer Bahadur Singh Purvanchal University, India

Keywords

Plant Disease Detection, Convolutional Neural Networks (CNN), Precision Agriculture, Artificial Intelligence