A Brief Comparative Study of YOLOv7, YOLOv8, and YOLOv9 for Threat Object Detection

Abstract

Threat object detection has become as essential task in X-ray security screening systems due to the rapid growth in population and possible dangerous threats in public spaces. Consequently, advanced Machine Learning (ML) and Deep Learning (DL) models have been utilized for object detection to mitigate the limitations found in traditional detection models such as slow training time, complex backgrounds, multi-scale prohibited objects, random stacking and occlusion scenarios. This paper introduces a comprehensive comparative study of three prominent detection models YOLOv7, YOLOv8, and YOLOv9 which achieved remarkable insights in real-time detection with notable performance and robustness. Experimental results conducted on the GDXray dataset show the outstanding performance of the three models, surpassing traditional models by achieving higher performance and deduction of false positives and false negatives rates.