AMOD-Net: Attention-based Multi-Scale Object Detection Network for X- Ray Baggage Security Inspection

Abstract

X-ray baggage security checking is an extremely important task, which can detect various dangerous objects in airports, stations and other public places to prevent crimes and protect personal safety. However, at present, most of the recognition is done manually, which is inefficient and error-prone. As a complementary, object detection algorithm is beneficial to avoiding errors caused by manual detection. Although the universal object detection is well developed and the performance of the universal detectors is very advanced, the performance of these detectors in X-ray image detection is mediocre. In this paper, we propose an Attention-based Multi-Scale Object Detection Network (called AMOD-Net) for X-ray baggage security inspection. To solve the problems of stacking and occlusion existed in the X- ray baggage image, we design a channel selection attention module for AMOD-Net. To make better use of the feature information, we construct a deep feature fusion structure for AMOD-Net. Experiments on the X-ray baggage dataset demonstrate that our approach achieves very competitive results.