Autoencoder and Modified YOLOv3 Based Firearms Object Detection in X-ray Baggage Images to Enhance Aviation Safety

Abstract

At airports and especially the baggage inspection task, the vital question that the human operator must answer is how to strike a balance between security screening, facilitation in a confined space, the good imypression of passengers through their passage, and speed of inspection. In order to help them reinvent their approach to control in such an environment, the help of automatic intelligent tools is necessary. This paper proposes firearms object detection based on modified YOLOv3 and autoencoder for security defense in dual X-ray images. The object detection is performed by a modified version of YOLOv3, to detect all the objects presented in the baggage. The object features are carried out by an autoencoder. The classification is performed by a Multi-Layer Perceptron (MLP) to classify a new object as a weapon or not. The proposed system has shown high efficiency in detecting firearms with a precision of 96.50%.