Detection and classification of dress code violations in educational environments using deep learning

Abstract

This paper explores the utilization of deep learning techniques for the detection and classification of dress code violations in educational environments, identifying the challenges of manual enforcement and the potential for systems that are automated. This paper exhibits a model that integrates Faster R-CNN for detection and EfficientNet for classification, which provides an accurate and very efficient system to monitor students’ compliance with the dress code policies. The model was trained on a dataset of images that were collected from Federal University Dutsin-Ma and were classified into “decent” and “indecent” dressing for both male and female students. The result achieved demonstrates that the model works efficiently, reaching a training accuracy of 98% and a validation accuracy of 96%, and with overall scores for precision, recall, and F1-score exceeding 97%, thereby proving its effectiveness in different dress code categories. The Uniformity across the techniques substantiates the feature extraction performance of the model and demonstrates its generalization ability. This paper outlines the benefits of automation in alleviating bias and human error by improving transparency and fairness and enforcing the dress code. The results showed how it is effective by combining powerful deep learning models with strong frameworks to solve problems of classification.