Spectral-based Inferential Measurement Of Grey System

On the basis of physical causality between light and matter and taking spectral measurement as the primary data space, spectral-based inferential measurement of grey system is an indirect measurement method by mapping the spectral information to the attribute dimensions with an empirical or semi-empirical mathematical model. In recent years, the demand for such measurements is increasing in the fields of lighting and display, agriculture and animal husbandry, food, biological medicine, etc. The core scientific issue of the paper is raised to calibration problem among different dimension scales. Essentially, it is a kind of dimension-reduction calibration by mapping the higher dimensional spectral data to the dimension of the interest variables. Aiming at this core scientific issue, a series research has beenworked on. Since the spectral data is a kind of numeric characteristic domain dependent on instruments and the inferential measurement is based on spectral data, the spectral-based inferential measurement is the indirect measurement technique including instrument and model. The research of this paper is carried out according to the following two main lines.Firstly, from the hardware point of view, the instrumental dependence of spectral data on wavelength and energy dimensions is studied by system optimization design of spectrometer; the matching relationship between the pixels sampling of detector and wavelength scale is studied by wavelength calibration; then then directional solid angle constraint based on light tube and hemispheric direction constraint based on cosine correction are studied; double disc coupling and direct coupling radiation transmission in irradiance calibration system are studied; the spatial geometrical features of diffuse light fromgrain samples are studied and a fiber bundle opticalcoupler withcircle-distributed for diffuse light collecting of grain sample is designed for matching that features. This main line is to illustrate the spectral measurement instrument as the scale system not only the data information dimensions are high, but also the parameters dimensions of hardware system for getting spectral domain information are relatively high. Different spectral measurement systems as characteristic scales of spectral domain information have much differences. Even though the design is completely same, there are scale deviations in the instrument for high parameters dimensions. So, spectral-based inferential measurement to the instrumental dependence influences the validity and consistency of model and the inferential measurement models made in this paper are only suitable for the hardware system used in this paper. The transfer of the models on hardware system will be an important research content in the future.Secondly, from the model point of view, the relationship and applicability of the mechanism model and data driven model are studied in the case of wavelength calibration; taking spectral irradiance calibration based on blackbody as example, model parameters influence and selection of large scale calibration is studied; based on radiation source temperature inferential measurement, when there are unknown parameters in the model, parameters fitting inferential method is studied by the advantage of high dimensions; based on color temperature inferential measurement of multi-primary color LED light source, multi-layer mapping model inferential measurement method is studied by taking spectral domain as intermediate mapping space; based on absorbance spectraof inferential measurement for protein components of wheat, the influences, recognition and processing of outliers in multivariate statistical calibration model of complex biological samples are studied; based on absorbance spectraof inferential measurement for rice taste, a perceptual inferential model in cross space mapping is also studied. The main problems and solving methods of different dimension scale calibration are studied through typical case. From the grey level of system, the unknown degree of system gets higher. Wavelength calibration and spectral irradiance calibration are clear mechanism models of the white system. But the mechanism models are much complex and difficult to be operated, it is considered as grey system, the polynomial fitting model is carried out for calibration; radiation source temperature inferential measurement is grey system with unknown parameters in the mechanism model; color temperature inferential measurement of multi-primary color LED light source belongs to high nonlinear grey system for optical electric heating multi-layer parameter mapping; the inferential measurement of wheat is grey system with solution space instability of complex biological samples; the inferential measurement for rice taste is mapping inference of grey system cross subject object space.The inferential measurement cases in this paper could achieve the requirement of measuring precision, and the validity of grey system is proved by spectra-based inferential measurement. This technique is able to build up empirical(or semi-empirical) model to realize different-dimension mapping measurement because of data driven calibration method. In different inferential measurement cases, the scale weight, parameters dimension weight and samples weight of calibration reference points in data driven calibration method are analyzed with leave one cross validation operator. The research results show that this kind of data driven method has the dependence on calibration reference points. In the case of limited calibration reference points, the subspace information analysis method is the only effective method for model validation and model optimization. From the contents of color temperature inferential measurement of LED light source and rice taste value inferential measurement, spectra-based inferential measurement is a technique which could establishes a cross-domain mapping model from spectral properties to the human physiological and psychological perception by calibration among different dimension scales. It is a human-based inferential measurement technique. With the further development of the technique, it will have broader application and development space.