A Thesis Submitted to the School of Post Graduate Studies in Partial Fulfillment of the Requirement for the Degree of Master of Science in Information Technolog

Abstract

Gamo community weaving represents cultural beauty and the unique elements of traditional clothing to symbolize the country, Ethiopia. The traditional weaving of Gamo are handcrafted goods that showcase the vibrant national culture, their indigenous knowledge and the art of creativity of the community. They have a rich history of creating their own weaving patterns and distinctive weaves for hand-woven fabrics. Experts and weavers alike are able to quickly identify and recognize differences in patterns. But most people have a hard time noticing the differences in it, especially young people and foreigners. The classification of these distinctive textiles is a skill that is frequently acquired through practice, which makes it an especially demanding and expensive form of learning and causes dissatisfaction among those who perform it. Digital archives, technology involvement, limited effort of studies to conduct on hand made patterns and use of machine learning techniques where more and diverse dataset are needed for enhanced classification using deep learning. Few authors conducted their studies on women’s dresses with embroidery, which is limited to specific areas, cultures, and design patterns; paper materials for Dunguza pattern classification and future references are also hard to come by for Gamo community weaving. Furthermore, teaching a machine to distinguish between these fabric patterns is the most challenging. However, the development of the widely used deep neural network algorithms for image classification presents a new opportunity to solve these issues with significantly better outcomes. This study introduces a novel approach for classifying between custom and original Dunguza patterns. Dunguza is a colorful hand-woven cloth which specifically represent the culture of the Gamo community. The study applied experimental research designs using a mixed research approach. To execute the main study tasks, such as data pre-processing, training, testing, and evaluating models, we used Python programming. For creating a conceptual model, we used E Draw Max. The study utilized a deep learning like CNN from scratch, combined CNN with canny edge detection algorithm and transfer learning where models trained on one task to be adapted for another related task. We employed CNN, VGG16, MobilNetV2, InceptionV3 and ResNet50 for our classification task. With this strategy, all of the models had an accuracy that surpassed 90%. Significantly, the ResNet50 model demonstrated a superior accuracy of 98% and the proposed CNN model comparatively attained a higher accuracy of 90%. Next, by combining the canny edge detection algorithm with this end-to-end CNN,80% accuracy registered. The finding of the experiment indicates that it would be beneficial to investigate the application of AI as a novel method of classifying weaving patterns in Ethiopia