Coffee Plant Disease Identification using Enhanced Short Learning EfficientNetV2

Plant diseases are prime and major root causes that directly impact agriculture production and quality. Reliable Classification and Identification of infection appearance at the initial stages could increase the yield and benefit the farmer economically. Hence an automotive system that can provide more reliable results in real-time is most needed for cash crops like coffee. In recent works, the Deep learning (DL) model and Machine Learning (ML) techniques have proven remarkable agricultural enhancements and improvements. Hence, significant work has been carried out by employing Deep Learning models, specifically with the help of the Transfer Learning (TL) design technique and Convolutional Neural Networks (CNN) for various crop species. This work addresses four different categories of coffee plant diseases. The proposed system avails the benefits of the EfficientNetV2-S TL model with slight modifications according to the dataset, namely Enhanced EfficientNetV2-S. The main idea of using the TL model is to enhance the training speed with fewer model parameters. Applying this technique with diverse model configurations at various training phases and varying image sizes, Enhanced EfficientNetV2-S outperformed with an accuracy of 98.1% compared to DenseNet-121 with 92.9% accuracy for 5000 images.