基于形态学和机器学习的固体核径迹图像优化识别算法  

Optimal recognition algorithm for solid nuclear track images based on morphology and machine learning

作　　者：张子扬 范胜男 李梦雪 周文珊[2] 邓君[1] ZHANG Ziyang;FAN Shengnan;LI Mengxue;ZHOU Wenshan;DENG Jun(National Institute for Radiological Protection,China Center for Disease Control and Prevention,Beijing 100088 China;Hubei Provincial Center for Disease Control and Prevention,Wuhan 430079 China)

机构地区：[1]中国疾病预防控制中心辐射防护与核安全医学所,北京100088 [2]湖北省疾病预防控制中心,湖北武汉430079

出　　处：《中国辐射卫生》2022年第3期290-295,共6页Chinese Journal of Radiological Health

摘　　要：目的基于机器学习方法,提出一种固体核径迹图像的计算机识别算法,实现核径迹的自动、快速和准确识别,提高固体径迹图像分析效率。方法首先利用形态学方法扫描143张含有径迹的图像,确定疑似径迹位置并截取1250张素材图。选取素材的50%为训练集、30%为验证集,训练机器学习模型。另选素材的20%为测试集,测试模型识别效果。算法代码基于MATLAB软件编写并训练。结果建立的固体径迹识别算法识别能力较强,测试集识别准确度可达84.8%。算法构建的机器学习模型程序能跟随训练数据量的投入不断进化,准确度进一步提升。结论本算法在图像形态学基础上结合机器学习对径迹识别算法进行了研究,较好地实现固体径迹的自动识别。未来将加大模型的数据投入,优化算法,提高识别准确度,以期为图像径迹自动识别提供更精确高效的方法。Objective To propose a computer recognition algorithm for solid nuclear track images based on the machine learning method,and to realize the automatic,fast and accurate recognition of nuclear tracks and improve the efficiency of solid track image analysis.Methods Firstly,143 images containing tracks were scanned by morphological method to de-termine the location of suspected tracks,and 1250 material images were captured.50%of the material were selected as the training set and 30%as the validation set for training the machine learning model.Another 20%of the material were selec-ted as the test set for testing the model recognition result.The algorithm code was written and trained based on the MAT-LAB software.Results The established solid track recognition algorithm had a strong recognition capability,and the re-cognition accuracy of the test set could reach 84.8%.The machine learning model program constructed by the algorithm could evolve continuously with the input of training data,further improving the accuracy.Conclusion Based on image morphology and machine learning,the track recognition algorithm was investigated,by which the automatic recognition of solid tracks was better realized.In the future,we will increase the data input of the model,optimize the algorithm,and im-prove the recognition accuracy,in order to provide a more accurate and efficient method for automatic image track recogni-tion.

关 键 词：固体核径迹 图像识别 机器学习 

分 类 号：TL815[核科学技术—核技术及应用]