Quantitative detection of turbid media components using textural features extracted from hyperspectral images

The accuracy of component detection of turbid media can be difficult to improve due to the mutual influence of scattering and absorption in light attenuation. In this study, a heteromorphic sample pool was introduced containing turbid media with India Ink and Intralipid-20% fat emulsion, which increases the scattering information of the non-circumferential symmetric hyperspectral image of the turbid media. A gray level co-oc-currence matrix (GLCM) was used to extract textural features from the hyperspectral images. Subsequently, the textural features were correlated with the concentrations of Intralipid-20% by means of partial least squares regression, and it was compared with the frequently used analysis of two-dimensional exit light intensity. Experimental results show that textural feature modeling is superior to conventional light intensity modeling with a correlation coefficient of prediction (Rp) = 0.9831 and a root-mean-square error of prediction (RMSEP) = 0.0631% in the prediction set. This study provides a potentially viable method for detecting the components of turbid media quantitatively in analytical chemistry.