【摘要】 X光违禁物品检测在保护公共社会安全中起着重要的作用，随着深度学习的发展，智能安检也发展迅速。针对违禁物品大小不一、物品之间相互遮挡等特点，本文提出一种改进Faster RCNN算法。该算法用具有更优图像特征提取特性的ResNeXt网络替换原来的VGG16网络，引入FPN网络以适应各种尺度的违禁物品，使用CIoU损失函数代替原来的SmoothL1损失函数。将改进后的Faster RCNN算法在OPIXray数据集上进行测试，实验结果表明，mAP值较原算法提升了12.4%,且对比当前主流目标检测框架YOLOv5、YOLOX的mAP值分别高出2%和2.2%。更多还原

【Abstract】 X-ray detection of prohibited goods plays an important role in protecting public and social security. With the development of deep learning, intelligent security inspection has also developed rapidly. This paper presents an improved Faster RCNN algorithm for the different sizes of prohibited articles and the mutual occlusion between goods. The original VGG16 network is replaced by the ResNeXt network with stronger image features, and the FPN network is introduced to adapt to prohibited items of various scales. Thereafter, the CIoU loss function is used to replace the original SmoothL1 loss function. The improved Faster RCNN algorithm is tested on the OPIXray datasets. The experimental results show that the mAP value is 12.4% higher than the original algorithm, and compared with the current mainstream object detection frameworks, the mAP values of YOLOv5 and YOLOX are 2% and 2.2% higher respectively.更多还原