基于深度学习的安检图像识别系统

Security Inspection Image Recognition System Based on Depth Learning

为快速准确的识别并定位安检图像中的危险品,本文对Faster RCNN算法进行改进提出一种实时检测并定位危险品的方法,并应用于安检机系统。根据目前的安检机的使用情况,设计并实现了基于深度学习的安检图像识别系统和安检设备客户端显示程序。系统主要设计思路为:准备数据集,应用Caffe框架搭建Faster RCNN模型,对模型进行训练得到一个识别率较高的模型,然后利用训练好的模型进行测试,输入安检图像,输出图像中危险物品的种类和位置。在安检机中安装基于深度学习的安检图像识别定位系统,通过网络连接把收到的安检图像传输到安检上位机,上位机调用安检图像识别定位系统完成物品的显示识别和定位。实验结果表明,系统拥有很好的实时检测的能力,尤其是小目标危险品的检测效果,极大地提高安检的效率。

In order to quickly and accurately identify and locate dangerous goods in the security image,this paper proposes a real-time detection and location method for dangerous goods based on Faster RCNN algorithm,and applies it to the security inspection machine system.According to the current use of security inspection machine,the security inspection image recognition system based on depth learning and the client display program of security inspection equipment are designed and implemented.The main design idea of the system is:prepare the data set,use Caffe framework to build Faster-RCNN model,train the model to get a high recognition rate model,then use the trained model to test,input the security image,output the type and location of dangerous goods in the image.Installed in the security machine based on depth learning security image recognition and positioning system,through the network connection received security image transmission to the security upper computer,the upper computer calls the security image recognition and positioning system to complete the display of goods recognition and positioning.Experimental results show that the system has a good real-time detection capability,especially the detection of small targets,greatly improving the efficiency of security inspection.