Building Deep Learning-Based Enset Leaf Disease Detection and Classification Model

                                       ABSTRACT The majority of Ethiopians still rely heavily on the agricultural sector, which continues to be one of the most significant in the country. Numerous serious issues affect this industry, including plant diseases that lower yield quality and productivity. One of the most important tasks in preventing the spread of various plant diseases is early detection and precise diagnosis of the health status of the plant, which should be done using technology rather than human labor. Enset crop is one of an Indigenous Ethiopian crop that sustains the most crucial agricultural products for Ethiopian people, businesses, and animals. It is the main portion of staple food for over 20 million people in Ethiopia's central, southern, and southwest. However, enset diseases have caused huge losses and reduced yields. Traditional observation methods by farmers or domain area experts may be time consuming, expensive, and sometimes inaccurate. The most accurate models for identified plant diseases are those based on deep learning techniques. One of the most common approaches to demonstrate computational models made up of several processing layers to learn representations of images with various levels of abstraction is the convolutional neural network (CNN). These models have dramatically improved the state-of-the-art in visual object recognition and image classification which makes it a good way to enset leaf disease detection and classification problems. The main purpose of this study is to design and develop a deep learning-based model for the identification and classification of the three most critical diseases of enset leaf such as enset bacterial wilt disease, fungal Leaf spot disease, and Virus Leaf streak disease. A total of 4932 images are identified, collected, prepared, and used for conducting experiments including augmented images with four different categories. More specially, this study focuses on three diseased Enset leaves and a healthy class obtained from the Southern Agricultural Research Institute in Hawassa, Areka Agricultural Research Center in Wolayita, and Sidama zone mersa Woredas these images are provided as input to the proposed model. the experimental results show that the proposed ResNet50 model can effectively detect and classify four classes of enset leaf diseases with the best classification accuracy of 98.75% which is higher than compared to other classical deep learning models such as VGG19, DenseNet201, InceptionV3, and CNN.7