X-DOG: An Intelligent X-ray-based Dangerous Goods Detection and Automatic Alarm System
Yu Shi, Yige Xu, Lai Wei, Haoran Gao, Xiaolong Xu · 2020
X-ray-based security inspection systems are widely used in public places, such as airports, train stations and other critical locations with a large-scale of crowds. However, security checks manually are not efficient enough. Current X-ray-based automatic detection methods, including deep learning and conventional ways, usually face low accuracy, poor universality, and other weakness. In this paper, we proposed an X-ray-based dangerous goods detection scheme named SSD-X, an improved SSD (Single Shot Multi-Box Detector) object detection algorithm, and built X-DOG: an intelligent X-ray-based dangerous goods detection and automatic alarm system with SSD-X. First, in consideration of the position uncertainty and the overlap of objects, multiple data enhancement is utilized to effectively improve the accuracy and ameliorate the overfitting phenomenon. To solve the problem of unbalanced positive and negative samples in detection algorithms, focal loss is adopted to the confidence loss function so as to accelerate the rate of convergence. What's more, soft-NMS is added to enhance performance of detection of dense objects. We built a portable detecting system X-DOG implemented with SSD-X, which can run on mobile platform. Real time detecting and automatic alarming functions are implemented in X-DOG. Compared with the baseline algorithm, SSD-X shows the better performance in our experiments.