| With the improvement of living standard over the years,consumers' demand for the quality and safety of agricultural and food products has been on steady increase.Accordingly,nondestructive quality detecting and automatic grading techniques and instruments have been developed.Thanks to the capability of measuring internal quality of products nondestructively,optical techniques including near-infrared spectroscopy(NIRS),multi-spectral imaging and hyper-spectral imaging are prevailing in the research area and have been put into applications extensively.These spectrum-based optical techniques correlate the internal quality indices of products with their spectra by chemometrics method,which is a black box approach.This statistical model lacks mathematical rigor since it is not based on a solid optical theory and measurement results are often dependent on instrument setup and some unkown variables.In addition,due to the lack of knowledge on how light interacts with medium,optical measurement instruments are set up mainly based on subjective experience,resulting in uncontrollable variables or parameters.It is,therefore,of great importance and necessity to measure the optical properties and study the light transport pattern in turbid media.This research combined Monte Carlo method,finite element method,hyperspectral imaging technique,spatially-resolved technique,integrating sphere method and data analysis and processing method to study the optical property measurement method with hyperspectral imaging-based spatially-resolved technique for horticultural products.Finite element simulations for light transport in turbid media were conducted using diffusion approximation to the radiative transfer equation(RTE)as control equaiton.The inverse algorithm for estimating the optical properties of tissues from spatially-resolved refletance was optimized to improve parameter estimating accuracy.A sequential estimation method was proposed and optimized for measuring the optical properties of two-layer media.A batch of solid phantoms mimicking the optical properties of horicultural products were fabricated for validating the measurement system and algorithm.The hyperspectral imaing-based system was calibrated and corrected in terms of spectral resolution,spatial resolution,and linearity.The optical properties of liquid phantoms,two-layer solid phantoms and real agro-products(mango and milk)were measured using the calibrated system and optimized parameter-extracting algorithm.This research was overall aimed at providing a nondestructive,accurate and fast approach to measure the optical properties of one-and two-layer horicultural products,which could provide valuable information on light transport in horticultural products,and also offer a solid theoretical base for optical detecting methods.The main research objectives,results and conclusions are as following:(1)Finite element method was applied to solve the diffusion approximation to the RTE with different boundary and illuminating conditions,through which the light transport in turbid media was simulated.The simulation results with infinite small beam as light source under three boundary conditions(Partial current boundary condition,PCBC;Extrapolated boundary condition,EBC;and Zero boundary condition,ZBC)showed that the PCBC and EBC performed similarly with Monte Carlo simulation in calculating fluence rate ?(z)within tissue,while ZBC yielded the worst result.Analytical solution to diffusion approximaiton and finite element method under PCBC and EBC provided lower surface fluence rate ?(r)and spatially-resolved diffuse reflectance R(r)compared to Monte Carlo simuations for small source-detector separations(r<1 mfp'),while for large source-detector separations(r>1 mfp'),analytical method,FEM-EBC and FEM-PCBC gave similar and good results matching well with that generated by Monte Carlo simulations.Results for finite size beams showed that FEM coupled with EBC could accurately simulate spatially-resolved diffuse reflectance at large distances(i.e.,when they were greated than the size of the beam).Large beam size resulted in larger error within the region of illumination.(2)The inverse algorithm for estimating the optical properties of agricultural and food products from spatially-resolved diffuse reflectance was optimized and a step-by-step parameter estimation method was proposed and validated.Results showed that:1)Curve fitting results using raw reflectance,smoothed data by moving average method and locally-weighted polynomial smoothing(loess)mehtod were similar without significant differecne at the 5%difference level,while results using data smoothed by robust-loess(rloess)was worse.2)Mean scaling normalization performed the best,compared with regular and smoothing normalization methods.3)The optimal numbers of data points for mean scaling normalization were optimized at 10,5 and 3 for spatial resolution of 0.05 mm,0.1 mm and 0.2 mm,respectively.4)The optimal start point decreased from 3 mfp's to 1.6 mfp's with the increase of ?s'/?a value from 10 to 60,while the recommended end point was kept at 16 mfp'.5)Compared with conventional 1-step method,the proposed step-by-step method improved parameter estimating accuracy significantly,with the estimate accuracy improving 18.4%and 51.4%for ?a and ?s',respectively.(3)A sequential method for measuring the optical properties of two-layer media with spatially-resolved diffuse reflectance was proposed,and the contraining conditions of the method were determined.In addition,factors affecting the estimation of the optical properties of top layer were analyzed.Results showed that:1)The constraining condition I was concluded as the thickness of top layer should be larger than two times mfp' when its optical properties were to be estimated using the diffusion model for homogeneous semi-infinite media.And the constraining condition ? was concluded as he thickness of top layer should be less than 16 times mfp' for accurately estimating the optical properties of the bottom layer.2)The term d/mfp1' had a significantly influece on the extraction of ?a1 and ?s1'.With the increse of d/mfp1' value,it became easier to extract ?a1 and ?s1' and the end point of optimal spatial region of reflectance increased.The effect of mfp1'/mfp2' and ?a1/?a2 can be represented by the term lieff1/?eff2.It was much easier to estimate ?a1 and ?s1' when ?eff1/?eff2 was close to 1.The term?s1'/?a1 did not have an obvious effect on the extraction of ?a1 and ?s1'.(4)The proposed sequential method for measuring the optical properties of two-layer media was optimized with Monte Carlo simulated data.For extracting ?a1 and ?s1',two rough estimating methods,'rough estimating method I' and 'rought estimating method II',were proposed.'Rough estimating method I' utilized the spatial region from 1 mm to 20 mm to extract?a1 and ?s1',while 'rough estimating method ?I' used 1.25 mm as the start point of spatial region,and the reflected intensity of 0.05 was used as threshold to determind the end point.Comparsion results showed that rough estimating method II performed better than method I in estimating ?a1 and ?s1'.The spatial region for accurate estimation of ?a1 and ?s1' was optimized in terms of mfp'.The start point of optimal spatial region was recommended at 1 mfp1',and the end point increased with the increase of d/mfp1' value.Based on the roughly estimated?a1 and ?s1' and the optimized spatial region,these two optical parameters were re-estimated.Results showed that the re-estimated ?a1 and ?s1',based on the rough results from 'rough estimating method I' were significantly improved,while those based on 'rought estimating method ?' kept at good accuracy.For extracting ?a2 and ?s27',the spatial region was also optimized on the basis that ?a1 and ?s1' were known prior.The start point of optimal spatial region was recommended at 3.5 mm,and the end point was determined by the reflected intensity of 5e-5.With this spatial region,?a2 and ?s2' were estimated using the sequential method.Results showed that the extracted ?a2 and ?s2' were significantly more accurate than those obtained by conventional 1-step method,with the average estimate accuracy improving 22%and 9%for ?a2 and ?s2',respectively.(5)A method for fabricating two-layer solid phantom was proposed and agro-product-mimicking phantoms were made.The optical properties of the fabricated phantoms was calibrated using integrating sphere system.Results showed that the optical properties and thickness of top layer could be accurately controlled,and the two layers could be perfectly attached to each other with this proposed method.(6)The optical properties of the fabricated two-layer phantoms,mango and milk samples within container were measured using the optical property analyzer(OPA)combined with the proposed sequential estimation method,and the error source was analyzed.Results showed that:1)OP A system combined with sequential estimation method could accurately measure the optical properties of top layer,with the average relative measurement errors for ?a1 less than 23%and for ?s1',less than 12%.2)The measured abosorbing spectra of mango peel was capable of presenting absorption peaks at certain wavelength regions,and the reduced scattering coefficients of mango peel were generally larger than that of flesh.3)The measured ?s' values of different concentrations of milk samples within container correlated highly with their concentration.The conefficients of determination R2 were 0.985 and 0.972 at the wavelengths of 630 nm and 850 nm,respectively.4)The unsatisfied measurement results of ?a2 were attributed to several factors,including systematic measurement errors of reflectance,inaccuracy of diffusion model,limitation of light source and dynamic range of camera,and the coupling effect of all four parameters to be measured. |
PDF Full Text Request