基于滤波器-机器学习的食欲抑制剂光谱信号预处理比较  

Comparison of spectral signal preprocessing of appetite inhibitors based on filter-machine learning

作　　者：古锟山 王继芬[1] 刘怡然 GU Kunshan;WANG Jifen;LIU Yiran(School of Investigation,People’s Public Security University of Chi‐na,Beijing 100038,China;School of Public Security Administration,People’s Public Security University of China,Beijing 100038,China)

机构地区：[1]中国人民公安大学侦查学院,北京100038 [2]中国人民公安大学公安管理学院,北京100038

出　　处：《分析试验室》2022年第5期539-546,共8页Chinese Journal of Analysis Laboratory

基　　金：中央高校基本科研业务费专项资金(2021JKF208)资助。

摘　　要：对现场缴获的食欲抑制剂进行快速检验能够为案件调查提供线索和方向,同时机器学习算法开展物证的快速无损检验是法庭物证学的重要研究之一。红外光谱是最经典的快速无损检验方法,滤波器能够有效地除去原始谱图的噪声和背景干扰,从而提高模型的识别效果。本文收集了从实际案件中缴获的4种食欲抑制剂样本共计291份,运用快速傅里叶变换滤波器和希尔伯特变换滤波器对样本原始光谱数据进行降噪处理,同时借助朴素贝叶斯和随机森林模型建立分类模型,开展识别工作,从而筛选除噪效果最优的滤波器,同时比较了朴素贝叶斯和随机森林模型的识别效果。结果表明,经滤波器处理后原始光谱数据的识别率和稳定性显著提升,希尔伯特变换滤波器的除噪效果要比快速傅里叶变换滤波器好,随机森林模型的识别率和稳定性均要比朴素贝叶斯模型强,随机森林模型对经希尔伯特变化滤波器处理后的训练集识别率为96.33%,测试集识别率为95.89%。该方法通过滤波器有效地滤除谱图的噪声,提高了模型定性识别能力,对法庭科学中食欲抑制剂的快速鉴定有一定的参考意义。The rapid test of on-site seized appetite inhibitors can provide clues and directions for case investigation.At the same time,the rapid non-destructive test of evidence by machine learning algorithm is one of the important studies of forensic evidence.Infrared spectroscopy is the most classic fast non-destructive testing method.Filter can effectively remove the noise and background interference of the original spectrum to improve the recognition effect of the model.Herein,a total of 291 samples of the 4 kinds of appetite inhibitors were collected from actual cases.Fast Flourier transform filter and Hilbert transform filter were used to denoise original spectral data.A classification model was established by Naive Bayes and Random Forest models,and the sample recognition was carried out to select filter with optimal denoising effect.At the same time,the recognition effect of Naive Bayes and Random Forest model was compared.The result showed that the recognition rate and stability of original spectral data improved significantly after filter processing.Denoising effect of Hilbert transform filter was better than that of Fast Fourier transform filter.Random Forest model was better than Naive Bayes model in recognition.The recognition rate of the Random Forest model for the training set processed by the Hilbert transform filter was 96.33%,and the test set was 95.89%.This method effectively filtered out the noise of the spectrum through the filter,and improved the qualitative identification ability of the model.It has certain reference significance for the rapid identification of appetite inhibitors in forensic science.

关 键 词：光谱学 食欲抑制剂 滤波器 朴素贝叶斯 随机森林 

分 类 号：O433.4[机械工程—光学工程] TN713[理学—光学]