Deep Learning-Based Predictive Analytics for Weapon Detection: A Fusion of FMR-CNN and YOLOv9 Models

The early detection of firearms from surveillance videos is necessary to improve public security and life. Even with the growing advancement of Deep Learning (DL) techniques, difficulties persist in distinguishing generic items from weapons in surveillance footage. While deep learning has significantly advanced generic object identification, weapons detection demands specialized methods that maintain high accuracy and processing efficiency. This study addresses these challenges by presenting a novel hybrid deep learning system that combines the strengths of Faster R-CNN, Mask R-CNN, and the YOLOv9s architecture. The aim is to enhance surveillance systems through predictive analytics and automatic weapon identification for defensive security measures. This research focuses on deep learning to create an intelligent detection system that improves operational effectiveness and recognition accuracy. The testing results indicate that the hybrid technique outperforms the individual models, achieving an overall accuracy of 99.02%, along with comparable precision and recall scores. In contrast, YOLOv9s alone achieved an accuracy of 98.70%, while other standalone models showed lower performance. However, the technique still faces limitations in poorly lit or visually cluttered environments.The proposed framework represents a flexible, scalable, and precise tool for instantaneous weapon spotting, contributing substantially to intelligent monitoring technologies and community protection efforts.