首页|基于改进Faster R-CNN的X光安检图像检测识别研究

基于改进Faster R-CNN的X光安检图像检测识别研究

扫码查看
针对X光图像危险品检测精度低的问题,提出一种基于改进Faster R-CNN的图像检测识别方法.通过在Faster R-CNN中增加一个预分类头部,并采用RoIAlign替代RoIPooling,避免RoIPooling阶段二次量化误差.结果表明,所提的改进Faster R-CNN检测方法,可有效实现对X光图像中的危险品检测.相较于标准Faster R-CNN与常用目标检测模型SSD、YOLOv3、RetinaNet,所提方法的检测准确率和F1分数值均得到不同程度的提升,分别达到92.65%和96.14,且具有更高的运行效率,识别正常图像的平均时长为0.01 s,识别异常图像的平均时长为0.18 s.
Research on X-ray Security Inspection Image Detection and Recognition Based on Improved Faster R-CNN
Aimed at the problem of low detection accuracy of dangerous goods in X-ray image,an image detection and recogni-tion method based on improved Faster R-CNN is proposed.A pre-classification header is added to Faster R-CNN and RoIAlign is used to replace the RoIPooling layer to avoid the secondary quantization error in the RoIPooling stage.The results show that the proposed improved Faster R-CNN detection method can effectively detect dangerous goods in X-ray images.Compared with the standard Faster R-CNN and the commonly used target detection models SSD,YOLOv3 and RetinaNet,the detection accu-racy and F1 score value of the proposed method are improved to varying degrees,reaching 92.65%and 96.14,respectively,and it has higher operation efficiency,the average recognition time of normal images is 0.01 s,and the average recognition time of abnormal images is 0.18 s.

deep learningX-ray imagedangerous goods inspectionFaster R-CNN

丁仲熙、钟昊、胡列峰

展开 >

长沙海关技术中心,湖南,长沙 410000

深度学习 X光图像 危险品检测 Faster R-CNN

2024

微型电脑应用
上海市微型电脑应用学会

微型电脑应用

CSTPCD
影响因子:0.359
ISSN:1007-757X
年,卷(期):2024.40(12)