X光安检图像多尺度违禁品检测

安检是保障人民生命财产安全的第一道防线,智能安检是安检行业未来发展的必然趋势。X光安检图像存在背景复杂、违禁品尺度多样以及相互遮挡现象,导致传统的目标检测算法无法获得满意的效果。本文在一阶段目标检测网络SSD框架的基础上,提出了适用于X光安检图像多尺度违禁品检测网络——非对称卷积多视野神经网络ACMNet(Asymmetrical Convolution Multi-View Neural Network)。检测网络增加了三个模块:小卷积非对称模块(Asymmetrical Tiny Convolution Module,ATM)、空洞多视野卷积模块(Dilated Convolution Multi-View Module,DCM)、多尺度特征图融合策略(Fusion strategy of multi-scale feature map,MF)。 ATM学习到的细节特征有助于小尺度违禁品的识别;DCM通过提供局部与全局之间的上下文特征信息来解决遮挡问题;MF则是通过融合高、低层特征图以提高模型在背景干扰情况下违禁品的检测精度。在仿真实验中,采用X光安检领域公开的数据集与自建的数据集,ACMNet在精确度上取得了令人满意的效果。 

Multi-scale prohibited item detection in X-ray security image

Security inspection is the first guarantee for the safety of people’s lives and property. Intelligent security inspection is an inevitable trend for the future development of the security inspection industry. On the consequences of the complex background, scales of prohibited items and blind with each other, traditional object detection algorithms always fail to get an excellent performance. Based on the object detection architecture of SSD neural network, we proposed a novel Asymmetrical Convolution Multi-View Neural Network(ACMNet), which gain a good result on multi-scale prohibited item detection in X-ray security image. Three modules have been added to the ACMNet: Asymmetrical Tiny Convolution Module (ATM), Dilated Convolution Multi-View Module(DCM), Fusion strategy of multi-scale feature map(MF). The detailed features learned by ATM module are going to identify small-scale prohibited item,the DCM module provides local and global information between different objects context to solve blind problems and MF strategy is performed to fuse feature maps of different scales between high and low level features to improve the accuracy of prohibited items under background interference in security images. In the simulation experiments, using the open source dataset and self-built data set in the field of X-ray security inspection, ACMNet has achieved satisfactory results in terms of accuracy.