基于改进加权聚类多特征融合的X射线图像识别研究  

Method for X-ray Image Detection Based on an Improved Weighted Clustering Multi-feature Fusion

作　　者：许玉婷 王强 XU Yu-ting;WANG Qiang(Chinese Academy of Customs Administration,Qinhuangdao 066004,China;School of Vehicles and Energy,Yanshan University,Qinhuangdao 066000,China;Institute of Nuclear and New Energy Technology,Tsinghua University,Beijing 100084,China;Beijing Key Laboratory on Nuclear Detection&Measurement Technology,Beijing 100084,China)

机构地区：[1]中国海关管理干部学院,秦皇岛066004 [2]燕山大学车辆与能源学院,秦皇岛066000 [3]清华大学核能与新能源技术研究院,北京100084 [4]核检测技术北京市重点实验室,北京100084

出　　处：《核电子学与探测技术》2025年第2期212-220,共9页Nuclear Electronics & Detection Technology

基　　金：海关总署科研项目“海关小样本辐射图像智能识别技术研究”(2023HK103);中国海关管理干部学院科研项目“智慧海关建设促进业务科技一体化研究”。

摘　　要：禁限类物品X射线图像的智能识别在维护安全方面具有重要意义,本文开展了小样本X射线图像智能识别技术研究。首先,提取了物品的傅里叶描述子(FD)、几何参数和不变矩作为原始特征。接着,提出了改进综合评价指标和加权聚类多特征融合方法,并对可调参数进行了理论和实验分析。最后,提出了基于改进加权聚类多特征融合和支持向量机的X射线图像识别方法。实验结果表明,改进综合评价指标的可调参数可以根据不同的识别目的和场景进行调整,实现识别最优化。对实际场景下的枪支和刀具X射线图像进行识别,准确率、召回率均在90%以上,虚警率低于10%,与其他识别方法相比,准确率高出9.67%,召回率高出25%。因此,本文提出的改进加权聚类多特征融合是一种有效的特征优化方法,基于改进加权聚类多特征融合和支持向量机的识别方法具备有效性、实时性和适用性的优点,该方法提供了一种X射线图像智能识别的关键技术。The intelligent detection of prohibited and restricted objects in X-ray images is of great significance in maintaining security.Therefore,the X-ray image detection with small samples is studied in this paper.Firstly,Fourier descriptors(FD),geometric parameters and invariant moments are extracted as original features.Then,an improved comprehensive evaluation index and the weighted clustering multi-feature fusion method are proposed,and adjustable parameters are analyzed by theory and experiments.Finally,the X-ray image detection method based on the weighted clustering multi-feature fusion and support vector machine is designed.The experimental results show that adjustable parameters of the improved comprehensive evaluation index can be adjusted for different detection purposes to achieve detection optimization.The actual X-ray images of guns and knives are detected with both accuracy and recall of higher than 90%,and the false alarm is lower than 10%.The accuracy exceeds 9.67%and the recall exceeds 25%than the other detection methods.Thus,the weighted clustering multi-feature fusion proposed in this paper is proved to be an effective feature optimization method.In addition,the X-ray image detection method based on weighted clustering multi-feature fusion and support vector machine has the advantages of validity,real-time and applicability.In conclusion,the method provides a key technology solution for X-ray image detection.

关 键 词：X射线成像 智能识别 综合评价指标 多特征融合 支持向量机 

分 类 号：TL822[核科学技术—核技术及应用]