YOLO MODEL DETECTION OF STUDENT NEATNESS BASED ON DEEP LEARNING: A SYSTEMTIC LITERATURE REVIEW

Abstract

Maintaining proper student neatness (uniform compliance, grooming standards, and posture) is essential for fostering disciplined learning environments. While traditional monitoring methods are labor-intensive and subjective, computer vision-based solutions leveraging You Only Look Once (YOLO) architectures offer promising alternatives. The objective of this study is to evaluate YOLO optimization techniques for student neatness detection, identify key challenges, and propose relevant future research directions. This systematic review evaluates 28 recent studies (2021-2024) to analyze optimization techniques for YOLO models in student neatness detection applications. Key findings demonstrate that attention-enhanced variants (e.g., YOLOv10-MSAM) achieve 87.0% mAP@0.5, while pruning and quantization methods enable real-time processing (50-130 FPS) on edge devices like Jetson Orin. The analysis reveals three critical challenges: (1) occlusion handling in crowded classrooms (10-15% false negatives), (2) lighting/background variability, and (3) ethical concerns regarding facial recognition. Emerging solutions include hybrid vision-language models for explainable detection and federated learning for privacy preservation. The review proposes a taxonomy of optimization approaches categorizing architectural modifications (attention mechanisms, lightweight backbones), data augmentation strategies (GAN-based synthesis), and deployment techniques (TensorRT acceleration). Future research directions emphasize multi-modal sensor fusion and domain adaptation for cross-institutional generalization. This work provides educators and AI developers with evidence-based guidelines for implementing automated neatness monitoring systems while addressing practical constraints in educational settings.