Research On Oil And Tumor Analysis Technology Based On Visiblenear Infrared Spectroscopy

With the advantages of non-contact,non-destructive and online rapid detection,spectroscopy technology has very important applications in many fields such as scientific research,military,industry,agriculture,and clinical medicine.This thesis mainly focuses on the visible-near-infrared spectroscopy detection technology in the analysis of the characteristics of high-iron lubricants and clinical tumor tissue labeling,and has developed a portable visible-near-infrared spectrum detection system and a simultaneous color and near-infrared fluorescence 3D The imaging bionic multi-mode 3D endoscopy system,and proposes the use of a neural network-based spectral classification algorithm for different applications,to achieve high-precision classification of oil and tumors,thus laying a good technical foundation for future online diagnosis.The main work of the thesis includes:First,in response to the portable non-contact rapid detection requirements of highspeed rail gearboxes,a portable visible-near-infrared spectrum detection system was developed to achieve ultra-wide visible-near-infrared spectrum detection from 330 nm to 1700 nm.This system is composed of light source,visible light spectrum module,infrared light spectrum module,data analysis circuit,information display screen,equipment power supply and data interface.At the same time,a reflection measurement probe is designed according to the structure of the observation window of the highspeed rail gearbox.The developed portable visible-near infrared spectroscopy detection system was used to quantitatively analyze the moisture content,viscosity,and particulate matter(iron,copper,silicon)content of high-speed rail gearbox lubricants.The transmission and reflection systems were used to collect spectra of different lubricating oil water content and modeled and analyzed.The prediction coefficients of transmission and reflection systems were 0.968 and 0.98312,and the standard prediction errors were 0.3128 and 0.2249,respectively.The prediction determination coefficient value and standard prediction error value of the reflection system are better than those of the transmission system.Secondly,the transmission spectrum data was collected on samples of different viscosity of lubricating oil.Compared with the BP neural network algorithm,the root means square error of the viscosity prediction results obtained by the quantum genetic-neural network algorithm was reduced from 0.3455 to 0.0294,and the coefficient of determination increased from 0.8504 to 0.9799.As a result,the predictive ability of the quantum genetic-neural network algorithm is significantly stronger than that of the BP neural network.Finally,the particulate matter in the lubricating oil,such as iron,copper,and silicon,was collected by reflectance spectroscopy data of single concentration and mixed concentration,and fitted with Monte Carlo-Extreme Learning regression,and finally obtained a single configuration of iron-containing lubricating oil,The predictive coefficients of determination for copper lubricating oil and silicon-containing lubricating oil are 0.9883,0.9974,and 0.9993 respectively,and the corresponding prediction root mean square errors are 0.00182,0.0048,and 0.0152,respectively.The prediction effect is very good.However,the prediction coefficients of determination for the mixed configuration of ironcontaining lubricating oil,copper-containing lubricating oil,and silicon-containing lubricating oil are 0.4969,0.7917,and 0.7558 respectively,and the corresponding predicted root mean square errors are 0.7075,0.7381,and 0.6850,respectively.The prediction results are not ideal,which shows that the detection of particulate matter by the portable visible and near infrared spectrometer system needs to be further improved.Subsequently,the hyperspectral ASD and the developed reflection probe were used to perform in vivo and in vitro spectroscopic detection of rabbit liver VX2 tumors and normal tissues of four rabbits,and then the support vector machine was used to achieve two classifications and four classifications respectively.The genetic algorithm is used to optimize the parameters of the support vector machine to study its classification.The results of the correction set and the prediction set of the two-class and four-class are both above 99%.The study found that as the number of variables decreases,the classification results can greatly reduce the running time of the algorithm while maintaining good accuracy.The results of the study have laid a solid foundation for the future multi-spectral and efficient real-time diagnosis of normal rabbit liver tissues and VX2 tumor tissues.Finally,based on the qualitative analysis of the tumor tissue spectrum,combined with clinical needs,a design of a biomimetic multi-mode 3D endoscope system that can simultaneously achieve color and near-infrared fluorescence 3D imaging is proposed.The system combines the characteristics of human eye binocular stereo imaging and compound eye multi-spectral imaging of bird-tailed mantis shrimp,and realizes simultaneous 3D imaging of red,green,blue,and near-infrared fluorescence.The prototype of the biomimetic multimode 3D endoscope has a working wavelength range of 400nm-1000 nm,including a dual-channel optical endoscope system,an optical relay system and a bionic multispectral sensor.By introducing an optical relay system,two images from the dual-channel optical endoscope system can be projected onto the same bionic multispectral sensor.The dual-channel optical endoscope system consists of a pair of hard mirrors.Each pair of hard mirrors includes a prism group with a 30-degree viewing angle,an imaging objective lens group and three pairs of image transmission cylinders.The bionic multi-mode 3D endoscope can provide doctors with real-time feedback on the position and size of tumor tissue in clinical surgical operations,to improve the success rate of the operation,reduce iatrogenic damage and reduce the recurrence rate.The innovations of the paper include:(1)In response to the demand for non-contact rapid detection of high-speed rail gearbox lubricants,a portable visible-near-infrared spectroscopy detection system was designed and developed,which can realize real-time online detection of gearbox lubricant components.(2)Using the developed portable visible-near infrared spectroscopy detection system,different regression and inversion algorithms were used to quantitatively analyze the viscosity,moisture content and particulate matter content of lubricating oil.(3)A qualitative study of rabbit tumor tissue was carried out using ASD and selfdeveloped reflection probe,and the classification accuracy of normal tissue and tumor tissue was over 99%.(4)By combining the advantages of human eye binocular stereo imaging and sparrow-tailed mantis shrimp multispectral imaging,a design scheme of bionic multimode 3D endoscope was proposed,and the corresponding principle prototype was developed.The prototype realizes three-dimensional,visible light color and nearinfrared fluorescence simultaneous imaging,providing doctors with real-time feedback on the position and size of the tumor tissue during surgery.