Semantic Segmentation Neural Network in Automatic Weapon Detection

Michal Wieczorek, Jakub Silka, Martyna Kobielnik, Marcin Wozniak

Abstract: The goal of this paper is to introduce a semantic segmentation neural network designed for the detection of firearms. The proposed network applies a fully convolutional architecture, incorporating features such as skip connections and batch normalization to enhance its performance. The network was trained using a vast dataset of annotated images and its performance was evaluated using a separate dataset. The results show that the proposed network is highly effective, achieving top-notch results in the detection of firearms. The network’s high accuracy 99.1% and ability to perform pixel-wise classification make it a valuable solution for real-world gun detection applications.