INDIGENOUS WEAPON DETECTION IN IMAGES FOR ENHANCED SECURITY IN NIGERIA USING YOLO APPROACH

This work explores the use of the ‘You Only Look Once’ (YOLO) deep learning model, for real-time weapon detection in images to enhance security. The work addresses the rising threat of insecurity of lives in Nigeria by developing an automated system capable of identifying guns and knives in public spaces, including the local indigenous ones. The primary dataset used consisted of publicly available foreign image datasets of handguns and knives. Also, a small custom dataset was created consisting of images of local indigenous weapons taken within Nigeria. These diverse datasets were pre-processed and used with YOLOv10 model for training, validation and test. Google Colab with GPU acceleration was used for the training. After validation, a mean average precision (mAP) of 64.4% was achieved for all the classes (57% for the knife class and 72% for the handgun class). On the test dataset, an overall precision of 76%, an overall recall of 64%, and an overall F1-score of 70% were achieved; thus, demonstrating some level of success in detecting local indigenous weapons, and detecting handguns better than knives, though challenges such as false positives, false negatives and misclassifications were noted. With larger custom dataset of images of local indigenous weapons for training, better results would be achieved. The work shows that the YOLO-based system can be integrated into existing security infrastructures of the country, with potential to improve existing security measures, and contributing to improved public safety.