Convolutional Neural Network for Nigerian Traditional Male Attire Classification
Umar Adam Ibrahim, Rukkaya Umar · 2021
Online business has brought about tremendous positive change globally; however, it comes with some challenges. Thousands of different products are posted on a daily basis. Thus, automatic tagging and classification are becoming extremely important for business environment. In effort to aid with classification problems, this paper finds a CNN structure with the minimal computational expense to classify male attires for three main ethnic groups in Nigeria; Yoruba, Igbo, and Hausa, while keeping prediction accuracy high. We tested several manually developed network structures and compared their accuracy with popular image classification networks, namely AlexNet. We collected a dataset of 1000 pictures, in 4 different classes. We achieved 72% of test accuracy on the shallow network built from scratch, performing better than AlexNet. The preliminary study shows that with proper image processing, a shallow network can be used for prediction, thus saving huge computational expense.