多色
胶水
胶粘剂
鉴定(生物学)
人工智能
物种鉴定
计算机科学
模式识别(心理学)
绘画
工程类
材料科学
机器学习
组分(热力学)
作者
Xinyu Xie,Wanru Wang,Qin Huang,Wenyuan Zhang,Xinyue Zhu,Xiaoyan (Jenny) Liu,Xinyue Zhu
出处
期刊:Analytical Methods
[Royal Society of Chemistry]
日期:2026-05-07
卷期号:18 (17): 3645-3656
摘要
principal component analysis (PCA), classification models developed based on support vector machines (SVMs), bayesian neural network (BNN), and random forest (RF) achieved cross-validated accuracies exceeding 91.8% for the four-species classification in unaged samples and >82.5% after 168 hours of UV-accelerated aging. To address the contact limitations of conventional ATR probes on non-planar surfaces, a polyethylene glycol (PEG)-based hydrogel sampling strategy was introduced, enabling the effective extraction of trace collagen signals from complex substrates. This method yielded above 86% accuracy in well-preserved samples. However, classification performance decreased significantly with extended aging, with severely degraded samples, such as those subjected to 336 hours of UV-accelerated aging, becoming indistinguishable. This work established a novel protocol that integrates minimally invasive sampling with machine learning for the species-specific identification of protein adhesives in cultural heritage. The proposed framework proves particularly effective for well-preserved relics, offering significant potential for advancing provenance studies and tailored conservation strategies.
科研通智能强力驱动
Strongly Powered by AbleSci AI