Multi-class classification and pretrained sustainable agriculture model for plant leaf disease detection using ResNet-101

In recent years, advancements in technology have greatly benefited the agricultural sector. One critical aspect of ensuring healthy plant growth is promptly identifying infected leaves. For sustainable agriculture, automatically detecting diseased on plant leaves before the infection spreads is highly advantageous. This paper introduces deep learning models like CNN, along with various optimization techniques and hyperparameter tuning being used. The study utilizes machine learning frameworks and a deep learning library. It encompasses 10 different datasets comprising 151,007 leaf images depicting 270 different diseases, including healthy leaves from 44 plant species. Additionally, pretrained models such as AlexNet, ResNet50, GoogLeNet, DenseNet-121, and ResNet101 are employed, with the ResNet model achieving the highest accuracy rate of 94.45%.