Multitask Deep Learning Approach for Habesha Fashion Cloth Recognition and Classification

Yohannes, Abenet

dc.contributor.author	Yohannes, Abenet
dc.date.accessioned	2024-04-19T08:23:21Z
dc.date.available	2024-04-19T08:23:21Z
dc.date.issued	2023-06
dc.identifier.uri	http://ir.bdu.edu.et/handle/123456789/15767
dc.description.abstract	Multitask learning is the most popular concept in deep learning, aiming to exploit the correlation among tasks. To achieve this, the learning of different tasks is performed jointly. This research proposes a deep learning-based multitask learning approach for Habesha fashion clothing recognition and fabric type classification. Previously, no research was conducted on Habesha fashion cloth recognition and classification by using multitasking learning. Currently, the demand for Habesha fashion cloth has increased for various formal events. Still, there is no mechanism in place to identify pure cultural clothing without using human vision. Not only does the technology that identify the class of clothes but there is also no standard platform for advertising such clothing across the globe using websites to increase the interaction between sellers and buyers. Therefore, this research aims to design and develop a multitask deep learning model for Habesha fashion clothes. The proposed model consists of two sub-networks working in tandem; the first identifies the fashion item from three Habesha fashion clothing categories: kemis, T-shirt, and Trouser and the second classifies the fabrics into five types: Abujede, Magg, Fasha, Mennen, and Shash. We have carried out experiments on 2322 image datasets collected from Bahir Dar, Gondar, and Addis Ababa Habesha fashion cloth shops to evaluate the performance of both sub-networks and achieve a 79.8% accuracy. We have also conducted experiments for the individual tasks to know the task which improves the performance of the model and we have got 94.6%, and 64.5% accuracy results for fashion type recognition and fabric type classification tasks respectively. The experimental results demonstrate that our proposed model achieves a promising result. When we train our model for the individual tasks we understood that our model was able to learn more distinctive features for differentiating between fashion types than between fabric types. This study’s main weakness is the lack of a qualified sufficient dataset to conduct an extensive experiment. Keywords: Multitask learning, Deep learning, Habesha Fashion Cloth	en_US
dc.language.iso	en_US	en_US
dc.subject	Software Engineering	en_US
dc.title	Multitask Deep Learning Approach for Habesha Fashion Cloth Recognition and Classification	en_US
dc.type	Thesis	en_US