Weapon Detection Using Region-Based Convolutional Neural Network (RCNN)

Automatic weapon detection in surveillance systems, increasing the safety of public areas. In this work, a deep-learningdetection method is presentedthat focuses on real-time performance. A multi-task loss function is used and an enhanced CNN architecture that improves the accuracy of localization and classification in our system. On a dataset of 50,000 images, the model shows excellent performance with a 99.38% IoU, 99.51% recall, 99.32% precision, and 99.47% accuracy. The model surpasses all the state-of-the-art methods, including SSD, YOLOv5, and Faster R-CNN, in terms of perseverance, inference speed, and detection accuracy. Its processing speed ensures real-time implementation of the system in surveillance scenarios as 158.64 FPS. Comparing the proposed system with others clearly shows the efficiency of the system in the management of diverse types of weapons with minimal false positives. These results set up a benchmark for real-time weapons detection with a high degree of accuracy, answering significant security issues