Detection and Classification of Ethiopian Kale Leaf Disease using Deep Learning

Ethiopian kale is one of the widely cultivable vegetables used for income and food consumption in raw and cooked ways. However, many factors, diseases, and pests affect production enhancement. Identification of Ethiopian kale disease is the main challenge in handling production loss, and it requires a huge amount of effort, cost, and time. To address this problem, researchers planned a deep learning approach for classifying Ethiopian kale diseases based on their leaves. To accomplish the proposed solution, the design science research methodology was followed. To conduct this study, a total of 5,372 images were collected from Areka, Gununo, and Boditi Woreda. The researchers applied image preprocessing techniques such as image resizing, background removal, normalization, and noise removal. A data augmentation technique was performed, and hyperparameter tuning was applied to get the optimum performance of the model. In this paper, five individual classes were studied, and from those classes, four are disease classes, namely black-rot, black-spot, downy-mildew, and ring-spot. Two convolutional neural network frameworks were applied, namely: the training deep neural network model from scratch and the transfer learning network model. Finally, the built classifier model has been evaluated through accuracy, precision, recall, and F1-score. Experimental results show that training accuracy obtained from transfer learning models InceptionV3, VGG19, and ResNet50 achieves an accuracy of 94.51%, 96.09%, and 74.54%, respectively. Conversely, the proposed model achieves an accuracy of 98.54%. As a result, a proposed model from scratch was selected for developing an effective and robust classifier model.