Deep Learning Based Gun Classification in X-Ray Images

In this study, deep learning based methods, which classifies the X-Ray images according to the presence of weapon, is proposed. In the first method the classification network has been trained with previously separated training images. In the second method, regions with weapons in the training images were cropped from the image. The training network was run with cropped images. In the third method, training was made with the regions determined in the images by Selective Search. In the last method, the training images were first given as an input image to the Faster R-CNN object detection method. Then, the regions determined as weapons were clipped from the image and included in the training set. Updated training set is given to the classification network. When the proposed four methods were compared, it was seen that the success was higher in the model in which the Faster R-CNN result was included in the training and test steps.