摘要
辐射成像是核辐射测量的一项重要内容。在辐射安全领域中典型的辐射成像方法主要包括针孔成像、(空间)编码孔径成像、时间编码成像和康普顿散射成像等,各有其优缺点与适用场景。基于现有成像技术方法上的积累以及先进成像设备研制经验,针对在“更快速、更准确、更大范围内”进行辐射探测(成像)的实际需要,本研究探讨了对辐射热点进行三维定位的三种可行性方案、人工智能在辐射热点动态成像、核素识别方面的应用以及辐射成像智能化平台。在辐射热点三维定位的研究方面,通过四目立体成像系统实现了远距离成像场景中放射源100 m距离下小于20%的测距精度,在室内场景中通过三维场景建模和辐射场构建实现了4枚放射源的三维定位。在人工智能算法的应用方面,实现了移动辐射热点的实时定位,可对25 m远处以36 km/h速度移动的14 mCi^(137)Cs放射源进行动态定位追踪。在神经网络核素识别算法研究方面,提出了一种基于先验能量信息的特征增强卷积神经网络,可在低计数和多核素混合的环境下,实现核素的快速识别。在辐射成像智能化平台研制方面,研制了可实现辐射热点检测的加速器隧道环境巡检机器人,并开发了一套车载式辐射应急监测平台,搭载有无人机载、车载、便携式γ相机等多个探测终端,可在辐射应急情况下实现城市区域的快速机动测量。本文所给出的诸多辐射成像智能化技术使得人们可以更快、更精确地获取放射源的空间三维分布与核素信息,为辐射环境评估、辐射防护、核应急演练和决策提供了先进高效的辐射探测技术手段。
Radiation imaging is an important aspect of nuclear radiation measurement.In the field of radiation safety,typical radiation imaging methods mainly include pinhole imaging,(spatial)coded aperture imaging,time-coded imaging,and Compton scattering imaging,etc,each with its own advantages,disadvantages,and applicable scenarios.Based on the accumulation of existing imaging technology methods and the experience in the development of advanced imaging equipment,this article explores three feasible schemes for the three-dimensional positioning of radiation hotspots,the application of artificial intelligence in the dynamic imaging of radiation hotspots,nuclear identification,and the intelligentization of radiation imaging platforms.It addresses the actual needs of radiation detection(imaging)that is“faster,more accurate,and covers a larger range.”In the research on the three-dimensional positioning of radiation hotspots,a four-eye stereo imaging system has achieved a distance measurement accuracy of less than 20%for radioactive sources at a distance of 100 meters in long-distance imaging scenarios.In indoor scenes,three-dimensional positioning of four radioactive sources has been achieved through three-dimensional scene modeling and radiation field construction.In the application of artificial intelligence algorithms,real-time positioning of moving radiation hotspots has been realized,capable of dynamically tracking and positioning a 14 mCi^(137)Cs radioactive source moving at a speed of 36 km/h at a distance of 25 meters.In the research on neural network nuclear identification algorithms,a feature-enhanced convolutional neural network based on prior energy information has been proposed,which can quickly identify nuclear substances in environments with low counts and mixed nuclear species.In the development of intelligent radiation imaging platforms,an accelerator tunnel environmental inspection robot capable of detecting radiation hotspots has been developed,and a vehicle-mounted radiation emergency monitoring platform has been created.Equipped with multiple detection terminals such as unmanned aerial vehicle(UAV)mounted,vehicle-mounted,and portable gamma cameras,it can quickly mobilize measurements in urban areas during radiation emergencies.The various intelligent radiation imaging technologies presented in this paper enable people to obtain the three-dimensional distribution and nuclear information of radioactive sources more quickly and accurately,providing advanced and efficient radiation detection technical means for radiation environmental assessment,radiation protection,nuclear emergency drills,and decision-making.
作者
魏龙
孔令钦
帅磊
梁秀佐
胡选侯
张译文
王晓明
WEI Long;KONG Lingqin;SHUAI Lei;LIANG Xiuzuo;HU Xuanhou;ZHANG Yiwen;WANG Xiaoming(Beijing Engineering Research Center of Radiographic Techniques and Equipment,Institute of High Energy Physics,Chinese Academy of Sciences,Beijing 100049,China;School of Nuclear Science and Technology,University of Chinese Academy of Sciences,Beijing 100049,China;Jinan Laboratory of Applied Nuclear Science,Jinan 250131,China;CAEA Center of Excellence on Nuclear Technology Applications for Nuclear Detection and Imaging,Beijing 100049,China)
出处
《原子核物理评论》
CAS
CSCD
北大核心
2024年第1期94-108,共15页
Nuclear Physics Review
基金
国家自然科学基金资助项目(12375307,12005234,12105116)。
关键词
核安全
辐射成像
智能化
nuclear safety
radiation imaging
intelligence
