K近邻算法结合红外光谱对轮胎橡胶颗粒的鉴别研究

详细信息    查看全文 | 推荐本文 |

英文篇名：Research on the Tire Rubber Based on K-Nearest Neighbor Algorithm and Infrared Spectroscopy 作者：桑国通 ; 廖晓曦 ; 何欣龙 ; 王继芬 英文作者：Sang Guotong;Liao Xiaoxi;He Xinlong;Wang Jifen;Institute of Forensic Science and Technology,People's Public Security University of China;School of International Police Studies,People's Public Security University of China; 关键词：红外光谱 ; 轮胎橡胶 ; K近邻算法 ; 种类鉴别 英文关键词：IR spectra;;Tire rubber;;KNN;;Identification 中文刊名：HXTB 英文刊名：Chemistry 机构：中国人民公安大学刑事科学技术学院;中国人民公安大学国际警务执法学院; 出版日期：2019-01-18 出版单位：化学通报 年：2019 期：v.82 语种：中文; 页：HXTB201901013 页数：5 CN：01 ISSN：11-1804/O6 分类号：90-94

摘要

在法庭科学领域,轮胎橡胶颗粒的检验鉴别对交通肇事和一些诉讼案件的侦破尤为重要,针对传统取样分析技术会破坏物证的问题和综合考察样本在多变量多维度上的差异性,提出基于红外光谱法结合K近邻算法无损识别轮胎橡胶的鉴别方法。采集不同品牌的样本,对其光谱进行自动基线校正和归一化操作,采用Savitsky-Golay算法平滑去噪,通过降维实现对840个原始特征到5个识别特征的高效筛选,运用训练样本为测试样本的方法进行交互验证,选取K值为1,"特征3"为主要自变量,"特征4"、"特征5"、"特征2"和"特征1"为协变量作为分类参数,按重要性加权特征进行计算样本之间的距离,建立分类模型,模型总分类准确率达83. 56%,区分效果良好,结合样本红外谱图展开进一步分析,最终成功将73类样本分为了10类。结果表明,利用红外光谱检测和K近邻算法可实现对轮胎橡胶颗粒的识别与分类,普适性和高效性较强,具有一定的借鉴和参考意义。
        In forensic science,it is particularly important for the rapid detection of traffic accidents and some lawsuits to identify the tire rubber. A method, non-destructive identification of tire rubber based on infrared spectroscopy and radial basis function neural network,is proposed based on the problem of traditional sampling and analysis techniques would destroy the physical evidence and the comprehensive consideration of multi-variable and multi-dimensional differences in samples. Samples of different brands were collected, and their spectra were automatically baseline corrected and normalized,and were smoothed using the Savitsky-Golay algorithm. Dimension reduction method is used to achieve efficient screening of 840 original features to 5 identification features. Using crossvalidation of training samples for test samples and selecting the K value of 1,"Feature 3"as the main independent variable,and"Feature 4","Feature 5","Feature 2",and"Feature 1"as covariates as classification parameters. A classification model was established through calculating the distance between samples according to the weighted feature of importance. The total classification accuracy rate is 83. 56%,and the differentiation effect is good. Finally,73 samples were successfully divided into 10 categories. The results showed that combing the IR spectral detection and KNN can identify the tire rubber spectra. The method has strong universality and high efficiency,and can provide certain reference significance.

引文

[1]张艳芬,孔波.中国石油和化工标准与质量,2016,36(3):3~5.
    [2]王震,董佳英.刑事技术,2014,(4):14~17.
    [3]刘艳,司民真,李伦等.光散射学报,2016,28(1):84~90.
    [4]杨永安,王登正,司民真.光散射学报,2015,27(4):379~383.
    [5]余静,张云,庞松颖等.光谱学与光谱分析,2016,36(9):2807~2811.
    [6]张金庄.辽宁警专学报,2014,16(3):59~63.
    [7]孟朝阳.光谱实验室,2009,26(6):1583~1586.
    [8]宋丽梅,罗菁.模式识别.北京:机械工业出版社,2015,82~85.
    [9]刘玲玲,武彦文,张旭等.化学学报,2012,70(8):995~1000.