远程LIBS结合GA-arPLS的爆炸物识别研究  

Explosive Residue Identification Study by Remote LIBS Combined With GA-arPLS

作　　者：闫红宇 赵宇[2] 陈媛媛[2] 刘昊 王志斌 YAN Hong-yu;ZHAO Yu;CHEN Yuan-yuan;LIU Hao;WANG Zhi-bin(School of Instrument and Electronics,North University of China,Tai yuan 030051,China;School of Information and Communication Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学仪器与电子学院,山西太原030051 [2]中北大学信息与通信工程学院,山西太原030051

出　　处：《光谱学与光谱分析》2024年第11期3199-3205,共7页Spectroscopy and Spectral Analysis

基　　金：山西省自然科学基金面上项目(202203021221103)资助。

摘　　要：为了保障公共安全和预防恐怖袭击事件的发生,提出基于遗传算法(GA)优化非对称重加权惩罚最小二乘(arPLS)的远程LIBS基线校正预处理方法,结合ANN分类模型实现6m距离下的四种爆炸物(TNT、RDX、HMX和CL-20)快速、准确识别。GA-arPLS算法基于arPLS引入适应度函数评估拟合基线,寻找候选参数空间中的最优解来实现拟合LIBS基线。由于LIBS光谱信号通常包括连续辐射、原子与分子发射线等噪声信息,其覆盖了LIBS光谱较宽的光波段;直接通过LIBS光谱对相似元素的有机物定性分析时,难以捕捉相似爆炸物的特征光谱之间微小差异实现分类,故远距离环境下通过GA-arPLS预处理来提高特征谱线辨识能力很有必要,因此提升光谱分析的准确度很有必要。研究将GA-arPLS校正前后的LIBS数据集分别作为支持向量机(SVM)和最邻近分类(KNN)的输入,SVM的分类准确率提升了8.4%,而KNN分类模型的准确率提升8.7%。分类准确率表明,该GA-arPLS基线校正预处理方法可有效降低远程LIBS光谱的连续背景,而结合人工神经网络(ANN)构建的分类模型对相似爆炸物的识别准确率从89.2%提升至100%,分类识别效果达到最优。研究表明,该基线校正预处理方法不仅有效减小远距离LIBS的连续背景辐射和噪声干扰,而且提升了远程LIBS分类模型的鲁棒性和预测能力。研究成果有望提升远程LIBS在爆炸物检测方面的准确性和效率,以更好地应对潜在的爆炸物威胁。This study proposes a remote LIBS baseline correction preprocessing method based on genetic algorithm(GA)optimized nonweighted penalty least squares(arPLS)to ensure public safety and prevent terrorist attacks.It combines this method with an ANN classification model to accurately identify four types of explosives(TNT,RDX,HMX,and CL-20)at a distance of 6 m.The GA-arPLS algorithm's foundation is adding a fitness function to arPLS,which allows it to assess the fitting baseline and choose the best option in the candidate parameter space for fitting the LIBS baseline.On the one hand,it is primarily caused by the instrument's inherent dark current noise,bremsstrahlung,or environmental factors.This is because LIBS spectral signals typically include noise signals such as continuous radiation and atomic and molecular emission lines,which cover a wide range of light bands in LIBS spectra.Therefore,in long-distance environments,it is necessary to improve the ability to identify characteristic spectral lines through GA-arPLS preprocessing;on the other hand,it is difficult to capture small differences between the characteristic spectra of similar explosives for classification when qualitatively analyzing organic compounds of similar elements directly through LIBS spectroscopy.As a result,spectral analysis accuracy needs to be raised.This study used the LIBS dataset as input for closest neighbor classification(KNN)and support vector machine(SVM)before and after GA-arPLS correction.SVM's classification accuracy increased by 8.4%,whereas the KNN model's accuracy increased by 8.7%.The classification accuracy demonstrates that the GA-arPLS baseline correction preprocessing method can effectively reduce the continuous background of remote LIBS spectra.Meanwhile,the artificial neural network(ANN)constructedclassification model achieves the optimal classification recognition effect by improving the recognition accuracy of similar explosives from 89.2%to 100%.Studies have demonstrated that this baseline correction preprocessing techniq

关 键 词：远程激光诱导击穿光谱 爆炸物识别 GA-arPLS预处理 人工神经网络(ANN) 基线校正 

分 类 号：O657.31[理学—分析化学]