Development of a coffee wilt disease identification model using deep learning

Abstract

Coffee, commonly known as Coffee Arabica, is Ethiopia's most significant beverage. Coffee production in Ethiopia is affected by a number of abiotic and biotic factors, the most prominent of which are diseases caused by a range of etiologic agents, especially fungus. A multitude of diseases harm the stems, leaves, fruits, and roots of the crop, lowering its yield and marketability. Ethiopia was the first place where the Fusarium wilt disease on Coffee Arabica was discovered. In this study, a deep learning method for the automatic detection of coffee wilt disease is proposed. The investigation was carried out in three stages. First, we collected images of healthy and diseased coffee. Secondly, we developed a convolutional neural network (CNN) that can tell the difference between healthy images and images of infected/diseased coffee leaves. Finally, we tuned the dataset to train and test the developed model. For the purpose of experimentation, we collected 4000 images of healthy and infected coffee leaves. 80% of the obtained dataset was utilized to train the model, while the remaining 20% is randomly picked to test the CNN model. The suggested model efficiently grouped the input image with a mean training accuracy of 98.1% and a mean test accuracy of 97.9% as a result of experimentation using a learning rate of 0.0001, a Sigmoid output layer activation function, 100 epochs, and an 8:2 training and testing dataset ratio.