基于深度学习的中子剂量率仪自动读数系统  

Development of an automatic recognition system for neutron dose meters based on deep learning

作　　者：王雨青 黄政林 孙博文 王桢 李胤 崔伟[1] 唐智辉 韦应靖[1,4] Wang Yuqing;Huang Zhenglin;Sun Bowen;Wang Zhen;Li Yin;Cui Wei;Tang Zhihui;Wei Yingjing(China Institute for Radiation Proctection,Shanxi Key Laboratory for Radiation Safety and Protection;Shanxi Key Laboratory for Radiation Safety and Protection,Taiyuan 030006,China;CNNC Key Laboratory for Radiation Protection Technology,Taiyuan 030006,China;Institute of Nuclear and New Energy Technology,Tsinghua University,Beijing 100084,China)

机构地区：[1]中国辐射防护研究院,山西太原030006 [2]辐射安全与防护山西省重点实验室,山西太原030006 [3]中核集团辐射防护技术重点实验室,山西太原030006 [4]清华大学核能与新能源技术研究院,北京100084

出　　处：《核电子学与探测技术》2025年第2期221-228,共8页Nuclear Electronics & Detection Technology

摘　　要：中子周围剂量当量监测是核电站、加速器等核设施周围辐射防护监测的重要内容。为提高中子周围剂量监测仪表检定的自动化程度,基于深度学习的文本检测与文本识别算法,建立了中子剂量率仪的自动读数系统。对比了三种不同的文本检测与文本识别模型对于中子剂量率仪表的识别效果,实验结果表明DBNet与CRNN模型具有较高的准确性且同时保证了推理的快速性。对20种常见的中子周围剂量当量率仪进行了识别测试,测试结果表明,自动读数系统对于所测试仪表的识别准确率均可达到90%以上,且经过滤算法后准确率可提升至100%,系统处理一帧图像的时间在0.7 s左右。Neutron surrounding dose monitoring is an important aspect of radiation monitoring around nuclear power plants,accelerators,and other nuclear facilities.To improve the automation level of neutron dose monitoring instrument verification,an automatic recognition system for neutron dose meters was established based on deep learning algorithms.Three different text detection and text recognition models were compared for the recognition performance.The experimental results show that DBNet and CRNN models have high accuracy while ensuring the speed of inference.A recognition test was conducted on 20 common neutron dose rate meters.The test results show that the automatic recognition system can achieve recognition accuracy of over 90%for the tested instruments,and the accuracy can be improved to 100%after filtering algorithms,and the processing time for one frame of image is about 0.7 seconds.

关 键 词：深度学习 中子周围剂量 计量检定 机器视觉 光学字符识别 自动化 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]