Research On Component Content Of Wedge-shaped Turbid Media Based On Hyperspectral Image

In order to overcome the shortcomings of the existing feature extraction methods of hyperspectral image to detect turbid media,this research has developed a new component content analysis method based on the wedged turbid media.This study combines near-infrared light technology,hyperspectral technology and chemometrics,and a rapid,non-invasive,low-cost,and accuracy method was provided to achieve the detection of the content of turbid media components.The main research contents of the paper include:(1)Use the optimized Monte Carlo wedge simulation model to establish a singlewavelength light propagation model of a turbid medium with different optical parameters,and analyze the three-dimensional light distribution of the wedge model slope,that is,the transmission surface.According to the simulation data,it is found that the light spot on the transmission surface of the wedge-shaped sample is oval-like and has non-circular symmetry.Based on this,a method for extracting characteristic regions of wedge-shaped turbid media with reduced scattering coefficients ranging from 1.5 to15 is proposed.This method can realize the simplification in the acquisition stage,realize the regression prediction of the optical parameters of the turbid medium by using several or more points,and according to the corresponding relationship between the optical parameters and the component concentration,it can be used to inversely infer the improvement of the turbidity The accuracy of the component content detection of the medium.(2)Designed and built a point-collection hyperspectral collection system based on fiber optic spectrometer.This system is different from the line array hyperspectral imaging system used in the previous research: first,the collection is more flexible,and the speed and accuracy of the collection can be adjusted as needed,and the smooth collection of spectral data is realized;the second is for uniform For solutions with no obvious spatial distribution,the collection sensitivity is higher,and it still has an excellent performance when collecting transmission spectra for high-concentration samples,which greatly increases the scope of applicable substances.This work provides a flexible and reliable experimental platform for later experiments.(3)Based on the experimental acquisition system built above,a simulation experiment was carried out to verify the feature region extraction method proposed in the simulation experiment.In the experiment,Intralipid,India Ink,and distilled water mixed in different proportions are configured to produce phantoms,which are in line with the range of optical parameters during simulation.A partial least squares regression model was established between the hyperspectral data of the extracted characteristic regions and the Intralipid concentration.Experimental results prove that the use of spectral data in characteristic regions to regress Intralipid concentration compared to modeling the overall spectral data not only reduces the complexity of the model,but also improves the detection accuracy.This topic makes the rapid,accurate and online detection of turbid media further possible.