Detection of coffee leaf disease and identification using deep learning

The sustainability of coffee production is a concern for many coffee-producing countries. Indeed, the conservation of the production of coffee needs to detect disease and infection in the early stages, to provide the ability to control and remedy. Coffee is one of the most consumed daily beverages, so it is considered one of the most important plant crops that affect the economy of the country that produces it. Thus, implementing systems for disease detection that do not require expert consultation can streamline production processes. In this paper, we proposed an efficient and rapid system utilizing cost-effective devices for detecting coffee leaf diseases to support farmers without the need for specialized expertise, leveraging deep learning models. Our technique involves several types of artificial intelligence (AI) models used, which include proposed new models using convolutional neural networks (CNN1, CNN2), and prompt transfer learning (VGG16, ResNet50, and EfficientNet), as well as applied machine learning supervised classifier with hybrid approaches (support vector machine (SVM) and Random Forest (RF)). In terms of training, the proposed model was fed with large datasets that contain five classes, with a total number of images exceeding 50,000 images, while the testing utilized a separate dataset. Finally, the results showed high performance across all evaluation metrics. CNN1 obviously distinguished that it has the superior accuracy compared to the other models, with a 99% value. In conclusion, the proposed framework has the capability to be applied, and it will deliver on-hand support to farmers.