预处理器
规范化(社会学)
多元统计
模式识别(心理学)
人工智能
计算机科学
平滑的
线性判别分析
鉴定(生物学)
统计
数学
计算机视觉
机器学习
植物
社会学
人类学
生物
作者
Qi Ge,Shihao Wu,Xiaolong Hou,Huichao Wang,Huaice Liu,Zhiyun Jia
标识
DOI:10.1016/j.microc.2023.109151
摘要
The inspection and identification of pigments is a critical task in forensic identification. Previously, investigators mainly analyzed them manually by comparing infrared spectra one by one, which was subject to significant subjective factors and time-consuming. This article proposes a non-destructive, fast, and accurate method for inspecting and identifying pigment evidence. In the experiment, 191 pigment samples from different brands were collected and analyzed for their infrared spectra. Multivariate scatter correction, Savitzky-Golay smoothing, and peak area normalization were used for preprocessing. Filter and multi-order derivative preprocessing were carried out, and feature extraction was performed using the CARS algorithm. Based on Fisher's discriminant, a classification model was established to distinguish between different brands of pigments accurately. The experimental results showed that infrared spectra combined with multivariate classification models could accurately identify pigment samples. The method was fast, non-destructive, and accurate, reducing inspection costs and improving efficiency. It can provide reference and guidance for inspecting and identifying relevant cases and other physical evidence inspections.
科研通智能强力驱动
Strongly Powered by AbleSci AI