The Evaluation of Deep Learning Using Convolutional Neural Network (CNN) Approach for Identifying Arabica and Robusta Coffee Plants

Abstract

Purpose Arabica and Robusta coffee plants are physically distinctive as manifested in their leaves, leaf shape, color, and size. However, for ordinary people or those who have just begun their business in coffee cultivation, identifying the type of coffee plant can be challenging. In this study, we incorporated and evaluated deep learning technology to identify the types of coffee based on leaf image identification. Methods In this study, we designed a deep learning architecture and compared it with the well-known approaches, including LeNet, AlexNet, ResNet-50, and GoogleNet. A total of 19,980 image datasets were split into training and testing data, consisting of 15,984 images and 3,996 images, respectively. Results The hyperparameters were taken into account where the use of 100 epoch and 0.0001 learning rate provided the highest accuracy. In addition, 10-fold cross-validation and ROC were used for evaluating the proposed architectures. The results show that the developed convolutional neural network (CNN) generated the highest accuracy of 97.67% compared to LeNet, AlexNet, ResNet-50, and GoogleNet with an accuracy rate of 97.20%, 95.10%, 72.35%, and 82,16%, respectively. Conclusions The modified-CNN algorithm had satisfactory accuracy in identifying different types of coffee. The underlying principles of such classification draw specific attention to the leaf shape, size, and color of Arabica and Robusta coffee. For future works, it is a potential method that can be used to rapidly identify diverse varieties of Robusta and Arabica coffee plants based on leaf tissue and above canopy characteristics.