摘要
通过高光谱成像技术,结合光谱数据特征提取、降维方法与机器学习,实现了对烟蒂物证的快速分类,为犯罪现场勘查提供了新的侦查方向和技术支撑。选取现场常见的不同品牌香烟为实验样本,提取烟丝的高光谱图像,每个型号取80个样本,共960个独立样品。在400~1040 nm光谱范围内,选取特定区域的反射率值作为光谱信息。采用蒙特卡洛-无变量信息消除(MC-UVE)、竞争性自适应重加权(CARS)和主成分分析(PCA)三种方法进行波段提取和数据降维,并结合反向传播神经网络(BPNN)和随机森林算法(RF),构建了6种优化模型。结果显示,PCA降维方法的检测性能最优,PCA-BP模型分类准确率高达92.1%,证明了高光谱成像技术在烟丝识别中的巨大应用潜力。
This study achieves rapid classification of cigarette butt evidence by utilizing hyperspectral imaging technology,combined with spectral data feature extraction,dimension reduction methods,and machine learning,providing new investigation directions and technical support for crime scene investigation.The study focuses on hyperspectral images of cigarette tobacco from different brands commonly found at crime scenes,with 80 samples taken from each type,totaling 960 independent samples.Within the spectral range of 400~1040 nm,the reflectance values of specific regions of interest(ROI)are selected as spectral information.Three methods,including Monte Carlo-Uninformative Variable Elimination(MC-UVE),Competitive Adaptive Reweighted Sampling(CARS),and Principal Component Analysis(PCA),are employed for band extraction and data dimension reduction.By combining these methods with Back Propagation Neural Network(BPNN)and Random Forest(RF)algorithms,six optimized models are constructed.The results indicate that the PCA dimension reduction method exhibits the best performance,with the PCA-BP model achieving a classification accuracy of up to 92.1%.This outcome demonstrates the significant potential of hyperspectral imaging technology in cigarette tobacco identification,providing a scientific basis and technical pathway for rapid and accurate identification of cigarette butt evidence at crime scenes.
作者
张宇帆
高树辉
ZHANG Yufan;GAO Shuhui(School of Investigation,People’s Public Security University of China,Beijing 100038,China)
出处
《应用化工》
北大核心
2025年第3期667-675,共9页
Applied Chemical Industry
基金
2023年度中央高校基本科研业务费团队建设学科基础理论体系和转化新课程研究项目(2023JKF01ZK11)。
关键词
高光谱成像
机器学习
特征提取
特征分类
烟丝
烟蒂物证
hyperspectral imaging
machine learning
feature extraction
feature classification
tobacco thread
cigarette butt evidence
