Comparative Evaluation of Yolo-Based Models for Multi-Class Weapon Detection

This study explores the effectiveness of various real-time object detectors for the problem of classifying different weapon types. To that end, the performance of YOLOv9, YOLOv10, YOLOv11, YOLOX, and YOLOR models were evaluated on a balanced dataset of 15,387 images across 5 weapon classes. All models are trained with 5-fold stratified cross-validation and YOLOR achieved the best overall performance with an mAP@0.5 of 0.8654 and mAP@0.5:0.95 of 0.8380. In addition, YOLOR obtained a precision of 0.8550, recall of 0.9570, and F1-score of 0.9210. Results show YOLOR’s suitability for high-precision, critical applications, while YOLOX offers advantages for recall-focused, resource-constrained deployments with a recall of 0.9628 despite a lower precision of 0.6902. The experiments highlighted the potential of real-time AI-based object detectors for enhancing security systems.