Higher YOLO with Consideration-Augmented Community and Enhanced Generalization Efficiency for Security Helmet Detection
Authors: Shuqi Shen, Junjie Yang
Summary: Security helmets play a vital position in defending staff from head accidents in development websites, the place potential hazards are prevalent. Nevertheless, presently, there isn’t any method that may concurrently obtain each mannequin accuracy and efficiency in advanced environments. On this research, we utilized a Yolo-based mannequin for security helmet detection, achieved a 2% enchancment in mAP (imply Common Precision) efficiency whereas lowering parameters and Flops depend by over 25%. YOLO(You Solely Look As soon as) is a extensively used, high-performance, light-weight mannequin structure that’s nicely fitted to advanced environments. We presents a novel method by incorporating a light-weight characteristic extraction community spine primarily based on GhostNetv2, integrating consideration modules equivalent to Spatial Channel-wise Consideration Web(SCNet) and Coordination Consideration Web(CANet), and adopting the Gradient Norm Conscious optimizer (GAM) for improved generalization means. In safety-critical environments, the correct detection and velocity of security helmets performs a pivotal position in stopping occupational hazards and making certain compliance with security protocols. This work addresses the urgent want for sturdy and environment friendly helmet detection strategies, providing a complete framework that not solely enhances accuracy but additionally improves the adaptability of detection fashions to real-world circumstances. Our experimental outcomes underscore the synergistic results of GhostNetv2, consideration modules, and the GAM optimizer, presenting a compelling answer for security helmet detection that achieves superior efficiency by way of accuracy, generalization, and effectivity