基于YOLO的违禁品检测深度卷积网络

A deep convolutional neural network based on YOLO for contraband detection

为自动识别X光图像中的违禁品,在YOLOv3深度卷积神经网络的基础上,文章针对多数违禁品对象尺寸较小,经常折叠交叉的特点,对YOLO网络的部分结构和机制进行了改进,构建了违禁品检测深度卷积神经网络探码器,用于对从安检系统获得的X光图像进行违禁品检测。实验结果表明,所提方法除了响应时间比YOLOv3稍长之外,在检测识别密集分布物体、小尺度目标及识别精度方面,均明显优于后者。

In order to automatically identify contrabands hiding in X-ray images,a new deep convolutional neural network,namely contraband detector which is used to detect contraband in the X-ray images obtained from the security inspection system,is constructed based on the YOLOv3.In the new network,some structures and mechanisms which originally come from YOLO network are significantly redesigned for detecting small-scale or overlapped contraband objects.The experimental results show that the performance of new network is better than YOLOv3 in terms of detection and recognition of densely distributed objects,small-scale targets and recognition accuracy,except for its slightly longer response time.