Classification of Coffee Leaves Diseases Using CNN

Abstract

Indonesia’s coffee is a one of major export and contributes significantly that generate foreign exchange to the country’s economy. The quality and quantity of coffee production depend on various factors such as humidity, rain, and fungus that can cause rust diseases on coffee leaves. This disease can spread quickly and affect other coffee plants quality, leading to decreased production. To address this issue, the CNN method with the VGG-19 architecture model was utilized to identify coffee plant diseases using image data and the python programming language, which in previous studies used MATLAB as their platform. In addition, VGG-19 has a more profound learning feature than the method used in previous studies, AlexNet which makes the structure of VGG-19 more detailed. The dataset that use in this paper is Robusta Coffee Leaf Images Dataset which have three classes and with 1560 images data in total, but only used 100 images in each classes. The VGG-19 model attained a precision level of 90% when the evaluated using the testing data with ratio 80:20, which 80% is training data, and 20% is validation data as testing data. This paper employed 0.0001 learning rate, batch size 15, momentum 0.9, 12 training iteration, and RMSprop optimizer.