| 基于深度学习的X光图像智能审像系统 |
投稿时间：2020-10-13 修订日期：2020-10-27  点此下载全文 |
| 引用本文： |
| 摘要点击次数: 153 |
| 全文下载次数: 0 |
|
| 基金项目:国家自然科学基金,国家自然科学基金项目（面上项目，重点项目，重大项目） |
|
| 中文摘要:针对X光安检机人工审核图片存在的效率低、误检和漏检等问题,设计并实现了一套基于Mask R-CNN算法的X光图片智能审像系统。实现了X光图像采集、数据汇聚、分析处理、违禁物品自动检测、数据存储等功能。通过分析比较,选择ResNet101作为BackBone训练网络,选取6000张X光图片作为样本,对刀、枪、液体瓶、手机、充电宝等五类违禁品进行标注。对训练参数优化调整,训练出违禁品的Mask R-CNN模型。在测试集上使用COCO评估方法,检出违禁品的平均精准率mAP50达到了0.83,明显高于Faster R-CNN、YOLOv3、SSD513等算法,具有实际工程应用价值。 |
| 中文关键词:深度学习 X光图像 目标检测 Mask R-CNN 违禁品检测 |
| |
| Intelligent X-ray Image Examination System Based on Deep Learning |
|
|
| Abstract:In order to solve the problems of low efficiency, false detection and missed inspection of X-ray security inspection machine, an intelligent X-ray image examination system based on Mask R-CNN algorithm is designed and implemented. Through analysis and comparison, ResNet101 is selected as the BackBone training network, and 6000 X-ray images are selected as samples to label five types of contraband such as knife, gun, liquid bottle, mobile phone and portable battery. The Mask R-CNN model of contraband is trained by optimizing the training parameters. Using COCO evaluation method in the test set, the average accuracy of detecting contraband is 0.83, which is significantly higher than Faster R-CNN、YOLOv3、SSD513 algorithm, which has practical engineering application value. |
| keywords:deep learning X-ray image object detection Mask R-CNN contraband detection |
| 查看全文 查看/发表评论 下载pdf阅读器 |
|
|
|
次