Object Identification and Localization of Visual Explanation for Weapon Detection

Abstract

Weapon detection is a difficult task that requires accurate identification of weapon objects in images. The object localization approach is mostly used because it combines a gradient with a convolutional layer to create a map of key locations on the image. This paper presents an Eigen-CAM method to localize and detect objects in an image for a Faster Region-based Convolutional Neural Network (Faster R-CNN) residual neural network (ResNet 50) model, giving a visual explanation. The Internet Movie Firearms Database (IMFDB) was used to train a deep learning model with the Faster R-CNN ResNet 50 model of the pre-trained PyTorch framework. Experimental results indicated that the Faster R-CNN ResNet 50 model achieved the highest mAP of 0.497 with 0.5 IoU. The Eigen-CAM method performed effectively for visual image representation.