Spectral Pattern Recognition of Cardiac Tissue in Electric Shock Death and Post-Mortem Electric Shock
LIU Xin-yu1, SHAO Wen-wu2*, ZHOU Shi-rui3
1. Key Laboratory of Microecology-immune Regulatory Network and Related Diseases, School of Basic Medicine, Jiamusi University, Jiamusi 154000, China
2. Jiamusi Infectious Disease Hospital, Jiamusi 154007, China
3. School of Criminology, People’s Public Security University of China, Beijing 100038, China
摘要： 在司法鉴定领域，涉及电击死亡的案件多发，鉴别死者是生前还是死后受到电击仍是法医病理学鉴定的难点问题之一。为此通过傅里叶变换红外光谱融合机器学习模型对心脏组织视角下的电击死和死后电击两种情况开展分类识别研究。将30只大鼠进行电击死、死后电击和对照处理，通过光谱仪扫描得到其心脏组织光谱，采用竞争性自适应重加权算法共提取到70个光谱特征波长，建立随机森林模型对特征波长提取前后的心脏组织光谱进行模式识别；结果表明，特征波长提取前后模型分类识别的准确率分别为34.9%和73.7%，验证了特征波长提取方法的有效性和必要性。同时建立偏最小二乘模型、传统支持向量机以及粒子群算法和灰狼算法优化的支持向量机模型进行分类识别，结果表明，模型分类识别的准确率分别为61.07%、34.48%、100%和98.46%，对比发现经特征提取后的粒子群优化支持向量机模型分类识别效果最好。为排除“生物学死亡期”的干扰，又取60只大鼠按同种方式对其处理，每组又分死后0.5 h和死后1 h 2个亚组，再次通过傅里叶变换红外光谱仪扫描得到光谱数据，数据预处理后将其与之前得到的数据进行一并处理并结合粒子群优化支持向量机模型分析，结果表明，该方法分类识别的准确率可达到80.85%。这为电击死领域的法医学鉴定提供了新的研究思路和方法，说明傅里叶红外变换红外光谱结合机器学习模型可以作为一种补充工具来提供相对客观的判断，具有重要的研究意义。
关键词：电击死；死后电击；特征波长；随机森林；分类识别；法医学鉴定
Abstract：In forensic identification, cases involving electrocution death are frequent and the identification of electric shock before and after death remains one of the difficult problems in forensic pathology identification. The experiments were carried out to classify and identify electrocution death and post-mortem electrocution behavior from the viewpoint of heart tissue through Fourier transform infrared spectroscopy fused with machine learning model. 30 rats were subjected to electrocution death, post-mortem electrocution and control treatment, and their heart tissue spectra were scanned by a spectrometer, and a total of 70 spectral feature wavelength points were extracted using a competitive adaptive re-weighting algorithm, and a random forest model was established to identify the feature wavelength The results showed that the accuracy of model classification recognition before and after feature wavelength extraction was 34.9% and 73.7% respectively, which verified the effectiveness and necessity of the feature wavelength extraction method, and the partial least squares model, traditional support vector machine and support vector machine model optimized by particle swarm algorithm and grey wolf algorithm were established for classification recognition. The results showed that the accuracy of the models was 61.07%, 34.48%, 100% and 98.46% respectively, and the particle swarm optimized support vector machine model with feature extraction was found to be the most effective. In order to exclude the interference of the “biological death period”, 60 rats were treated in the same way, and each group was divided into two subgroups: 0.5 h and 1 h post-mortem, and the spectral data were scanned again by Fourier transform infrared spectrometer-SVM model analysis, the results showed that the method could achieve an accuracy of 80.85% for classification and identification. It provides a new research idea and method for forensic identification in electrocution death. It shows that FTIR combined with machine learning models can be an essential research significance as a complementary tool to provide relatively objective judgements.
刘昕宇，邵文武，周世瑞. 电击死与死后电击的心脏组织光谱模式识别[J]. 光谱学与光谱分析, 2023, 43(04): 1126-1133.
LIU Xin-yu, SHAO Wen-wu, ZHOU Shi-rui. Spectral Pattern Recognition of Cardiac Tissue in Electric Shock Death and Post-Mortem Electric Shock. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2023, 43(04): 1126-1133.
[1] Byard R W, Thu M, Gilbert J D. Forensic Science, Medicine, and Pathology, 2017, 13(3): 390.
[2] FENG Guo-wei, LIU Xia, QI Qian, et al(冯国伟, 刘 霞, 齐 倩, 等). Journal of Forensic Medicine(法医学杂志), 2021, 37(2): 158.
[3] ZHANG Guo-zhong, LI Rui-li, FENG Guo-wei, et al(张国忠, 李瑞利, 冯国伟, 等). Journal of Forensic Medicine(法医学杂志), 2015, 31(4): 262.
[4] Palmiere C, Teresiński G, Hejna P. International Journal of Legal Medicine, 2014, 128(4): 607.
[5] Li S Y, Zou D H, Luo Y W, et al. Science & Justice, 2014, 54(1): 98.
[6] Zhang J, Lin W, Lin H, et al. PloS one, 2017, 12(1): e0170844.
[7] Lin H, Luo Y, Sun Q, et al. Analytical Chemistry, 2018, 90(4): 2708.
[8] DONG Chao-xiu, DONG Zu-xin, ZHANG Gao-qin, et al(董超秀, 董祖鑫, 张高勤, 等). Chinese Journal of Forensic Medicine(中国法医学杂志), 2020, 35(5): 479.
[9] Shetty B S K, Kanchan T, Acharya J, et al. Burns, 2014, 40(7): e45.
[10] ZHANG Guo-zhong, LI Rui-li, FENG Guo-wei, et al(张国忠，李瑞利，冯国伟，等). Journal of Forensic Medicine(法医学杂志), 2015, 31(4): 262.
[11] Wu D, Luo Y W, Zhang J, et al. Spectrochim Acta A Mol Biomol Spectrosc，2021，258: 119882.
[12] LUO Yi-jia, ZHU He, LI Xiao-han, et al(罗一甲, 祝 赫, 李潇涵, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2021, 41(7): 2036.
[13] XIONG Xin, SUN Zhong-chun, QU Wei, et al(熊 鑫, 孙仲春, 瞿 巍，等). Chinese Journal of Forensic Medicine(中国法医学杂志), 2020, 35(4): 403.
[14] Trevisan J, Angelov P P, Patel I I, et al. Analyst, 2010, 135(12): 3266.
[15] Leskovar T, Pajnič I Z, Geršak Ž M, et al. Forensic Science International: Genetics，2020, 44: 102196.
[16] Guleken Z, Jakubczyk P, Wiesław P, et al. Talanta，2022，237: 122916.
[17] DONG Na-na, CHEN Xiao-lan, DENG Bi-li, et al(董娜娜, 陈晓兰, 邓铋莉, 等). China Modern Applied Pharmacy(中国现代应用药学), 2022, 39(2): 141.
[18] SU Cheng-ke, LIU Cui-mei, MENG Xin, et al(苏承轲, 刘翠梅, 孟 鑫, 等). Journal of Forensic Medicine(法医学杂志), 2021, 37(1): 33.
[19] WANG Xiao-liang, ZHANG Bing-hua, DU Wei, et al(王小亮, 张秉华, 杜 玮, 等). Chinese Pharmacist(中国药师), 2021, 24(8): 571.
[20] SHI Wei-guo, SONG Cun-li(时维国, 宋存利). Computer Integrated Manufacturing System(计算机集成制造系统), 2021, 27(11): 3196.
[21] ZHAO Yu-xuan, HUANG Wei, DONG Lin-pei, et al(赵昱萱, 黄 威, 董林沛, 等). Chemical Research and Application(化学研究与应用), 2022, 34(1): 91.
