Research On Hyperspectral Image Analysis Methods For Wedge-shaped Samples

The rapid non-destructive detection of turbid media has always been an important research content in the fields of food,medicine,and environment.However,due to the complex composition of the turbid media and the existence of absorption and scattering effects at the same time,it is difficult to improve the detection accuracy of the turbid medium component content when it is analyzed by spectral analysis technology.Studies have shown that the hyperspectral images of wedge samples can obtain more dimensional information than the traditional hyperspectral images of flat samples,but the extraction methods of effective information related to the content of components to be measured in hyperspectral images need to be further studied.The main content of this research includes the following three aspects.(1)In order to study the propagation law of light in wedge sample,the propagation model of photon in turbid wedge sample was established.Based on the traditional Monte Carlo simulation model of flat plate,the Monte Carlo simulation model of photon propagation in wedge sample was established by modifying the boundary conditions,and the properties of wedge model were studied.In order to solve the problem of too long calculation time of wedge model simulation,the model was optimized.The calculation time of the optimized model is 1/30 of the original model.(2)In order to analyze the influence of the optical parameters of the sample on the light intensity distribution of the wedge sample transmission surface,this study extracted the characteristic information related to the content of the component to be measured from the light intensity distribution map of the wedge sample transmission surface.The control variable method was used to modify the optical parameters to obtain the light intensity distribution on the transmission surface of the wedge sample under different optical parameters.The information related to the content of the components to be measured was extracted from the light intensity distribution image,and the change rule of the characteristic information with the optical parameters was studied.Two hyperspectral image analysis methods for wedge samples were proposed:multi-dimensional feature parameter method and multi-dimensional radial distance method.(3)In order to verify the effectiveness of multi-dimensional characteristic parameter method and multi-dimensional radial distance method in improving the detection accuracy of turbid media,a hyperspectral image acquisition system based on wedge sample was studied and designed.The mixture of Intralipid-20% and India ink was used as the turbid medium to be measured.The effectiveness of the multi-dimensional characteristic parameter method and the multidimensional radial distance method in improving the detection accuracy of turbid media was verified by modeling and analyzing the two methods respectively.The results showed that extracting the information related to the content of the components to be measured in hyperspectral images,and using the feature information modeling analysis,can enhance the correlation between the spectral characteristics of the modeling data and the content of the component to be measured,and effectively improve the detection accuracy of turbid media.The research method in this paper provides an efficient and non-invasive chemical composition analysis method for the detection of turbid media using near-infrared spectroscopy,and has certain guiding significance for the quantitative composition analysis of turbid media.