1. Identification accuracy improvement of non-uniform paper samples
- Author
-
Yoshihiko Nakayama, Hideya Taniguchi, Shigeru Sugawara, Ichiro Ishimaru, and Natsumi Kawashima
- Subjects
business.industry ,Hyperspectral imaging ,Sample (statistics) ,Pattern recognition ,02 engineering and technology ,021001 nanoscience & nanotechnology ,Condensed Matter Physics ,Accuracy improvement ,01 natural sciences ,Measure (mathematics) ,Atomic and Molecular Physics, and Optics ,Spectral line ,Electronic, Optical and Magnetic Materials ,010309 optics ,Identification (information) ,0103 physical sciences ,Artificial intelligence ,0210 nano-technology ,business ,Mathematics - Abstract
Paper is non-uniform; thus, it is difficult to ascertain whether the slight differences in near-infrared absorption spectra are caused by the difference in paper type or the paper’s non-uniformity. We used near-infrared hyperspectral imaging to measure the spectra of many different points quickly for eight different paper samples. Next, we analysed the data using machine learning and constructed a model to ascertain whether the differences in the spectra are due to the difference in paper type or the paper’s non-uniformity. Even papers with very similar spectral shapes were distinguished with high accuracy with this method. As the number of spectra used to construct the classification model increased, the accuracy of the classification model increased. With 6840 spectra per sample, the accuracy exceeded 85%, and with 68,400 spectra per sample, the accuracy for identification was 87.8%.
- Published
- 2019
- Full Text
- View/download PDF