Deep Learning Based Enset Leaf Disease Classification Model In The Case of Ethiopia

Abstract

Ethiopia is one of the developing countries in the world, and its economy depends on both agriculture and industry. From agriculture products, Enset crop is one of the most important for human beings, industries as well as for animals in Ethiopia. Enset is one of such crops, and it is consumed as a staple food by around 20 million people in Ethiopia's center, south, and southwest areas. It is very susceptible to different types of diseases which is caused by bacteria, fungi, and virus harm the enset crop. Among this diseases, enset bacterial wilt, enset leaf spot and enset leaf streak are common diseases that attack enset leaf. It is impossible for plant pathologists to access each and every enset crop to notice those illnesses. There are traditional mechanisms to classify enset leaf diseases by visual observation. However, traditional mechanisms for classifying enset leaf diseases have drawbacks such as being expensive, inconsistent and prone to error, taking more time, requiring professional staff, specialized instruments, inefficient, and so on. As a result, we are motivated to create an enset leaf disease classification model using deep learning techniques to assist experts. Sample of enset leaf disease images were taken from Areka Agricultural research center. It is proposed to classify enset leaf diseases using deep learning model. The proposed approach has two main phases. In first phase the designed model is trained and tested by collected dataset and classify the enset leaf disease using by different neural network. Finally, the deep learning model that can classify the given image in to health enset leaf, bacterial wilt disease, leaf spot disease and leaf streak disease is done. The dataset contains 1814 original Enset images. From this, 80% of the images are used for training and the rest for testing the model. During training, data augmentation technique is used to generate more images to fit the proposed model using by Keras libraries. The Convolutional Neural Network, Multilayer perceptron and Hybrid of Convolutional Neural Network and Long Short Term Memory model can successfully classify the given image with an accuracy of 96.81%, 75%, and 94.83% respectively.