A Systematized Chronicity based Disease Classification in Coffee Leaves using Deep Learning

Diseases have an impact on agriculture production and quality, which is essential to global economic prosperity and growth. For agriculture to be sustainable over the long run disease diagnosis on the plant is important. It is difficult to monitor plant diseases manually because of time limitations and the multiplicity of diseases. The integration of computer vision and deep learning algorithms enables the quick and precise identification of disease causes. Therefore, these issues may be reduced as quickly as feasible by applying remedial measures. This research includes an automated convolutional neural network (CNN) technique for categorizing coffee leaf diseases that are based on deep learning. In order to strengthen and improve the system data augmentation approach is used. The proposed model is trained using max pooling, dropout layer, and with two dense layers and Adam optimizer on the dataset of arabica coffee leaves in various states, such as Healthy, Miner, Rust, Phoma, and Cescospora. The trained model obtained an accuracy of 99.70% and 99.84% on unseen images in balance datasets for 3-class and 5-class, respectively whereas it achieved an accuracy of 98.60%, and 97.50% on unseen test datasets for 3-class and 5-class, respectively for unbalanced datasets. When compared to the trained single network, performance metrics showed that the proposed model produced more precise and better results.