Improving public safety through automatic firearm detection and categorization: leveraging transfer learning and mask R-CNN for accurate handgun identification and classification

Abstract

The new emerging portable weapons, such as firearms, pistols, and revolvers, used in the commission of a crime, highlight the need for improved surveillance systems that can assist in recognizing such behavioral patterns to discourage crime. In most cases, surveillance is done manually, which creates many opportunities for mistakes and consumes much time. The subtype of AI, particularly in object recognition, classification, and picture segmentation, known as deep learning, can offer a potential solution in weapon detection. All these tasks can now be solved using Convolutional Neural Networks (CNNs); some of the most outstanding models in the market today, such as Faster R-CNN, YOLO, and Mask R-CNN, are accurate and real-time. In our study, Mask R-CNN demonstrated the highest accuracy and stability in handgun detection, even under challenging conditions. The model achieved an IOU of 0.7901 and a Dice score of 0.8828, highlighting its effectiveness in accurately identifying firearms. Compared to Faster R-CNN and YOLO, it showed superior performance in segmenting objects with greater precision. The evaluation metrics, including Precision, Recall, and F1-Score, further confirmed the effectiveness of Mask R-CNN, with an F1-Score of 1.0 in the detection task. All the above-mentioned deep learning models are explained in this paper, with a special focus on the results of the Mask R-CNN model as a portable firearms detector. The results also suggest improvements in transfer learning and fine-tuning pre-trained architectures to optimize weapon detection. Thus, the goal of the present work is to suggest an enhanced automatic firearm detection system that will increase security in public areas by nearly eliminating the human factor. This paper employs a methodical approach to the development and overall evaluation of the system, providing practical recommendations for creating weapon recognition systems in the real world.