摘要： 为了快速识别市场常见的防火涂料品牌，结合光谱成像与机器学习，提出了 2 种快速检测防火涂料一致性的方法。采用高光谱成像和短视频成像技术，测量了 7 种品牌防火涂料样品的光谱，利用主成分分析法对光谱数据进行降维，表明各品牌存在可分性。对光谱数据进行预处理、划分训练集和测试集后，评估常用机器学习方法的分类准确度，包括最小二乘判别分析、支持向量机等。结果表明：将光谱成像技术与机器学习结合，能够准确地区分防火涂料的品牌。短视频成像仅需智能手机即可实现光谱采集，具有技术成本低、操作便捷等优势，该技术与机器学习结合，在现场原位检测防火涂料的一致性有更广阔的应用前景。

关键词: 防火涂料, 高光谱成像, 短视频成像, 机器学习, 分类

Abstract: In order to quickly identify common fire retardant coatings brands in the market, this paper combined spectral imaging and machine learning to propose two methods to quickly detect the consistency of fire retardant coatings. Hyperspectral imaging and short video imaging technology were used to measure the spectra of seven brands of fire retardant coating samples, and the spectral data was reduced by principal component analysis to indicate the separability between samples of each brand. After preprocessing the spectral data, dividing the training and test sets, the classification accuracy of common machine learning methods, including least squares discriminant analysis and support vector machines, was evaluated. The results showed that the combination of spectral imaging technology and machine learning can accurately distinguish fire retardant coating brands. Short video imaging only required smart phones to achieve spectral acquisition, which had the advantages of low technical cost and convenient operation. The combination of this technology and machine learning had broader application prospects for in situ testing of the consistency of fire retardant coatings on site.

Key words: fire retardant coatings, hyperspectral imaging, short video imaging, machine learning, classification

中图分类号: 

• TQ637.6