Study On The Inversion Model Of Optical Properties For Fruit Tissue And Its Application

Optical properties of fruit are important parameter to reflect itself chemical composition,physical structure, physiological, pathological state. Measurement of optical properties of the fruit has important significance for studying imaging of the tissue internal structure and photon transmission characteristics, analyzing of chemical properties and physical structure,establishing detection and evaluation model of tissue internal quality/state. This paper utilized the MC simulation to study optical transmission characteristics in single fruit tissue, and solve optical properties. Then,the tissue simulating liquid images were acquisition using steady-state spatially resolved hyperspectral scattering imaging technique. The optical properties of tissue simulating liquid can be solved combining with the developed nonlinear inversion regression model. On the basis of the research results, the absorption and scattering spectra of apple were studied, and the prediction models were established between the spectral characteristics and the hardness and soluble solids content. The main contents of this paper are as follows:Diffusion model has a disadvantage that larger deviations will exist between the model and MC simulation if the detector is close to the light source. Therefore, the article proposed a new method based on the iterative inversion estimation of transport mean free path and the minimum source-detector distance determination. The method utilizes the idea of iterative estimation and adaptively evaluates the value of transport mean free path and changes the minimum source-detector distance to acquire a reasonable data interval for the inversion of optical properties. The results showed that compared to the traditional empirical estimation method, the iterative inversion method can reduce the error of the near light source and effectively improve the inversion accuracy of the optical properties of the fruit tissue. Under the condition of no noise, the average relative error of the absorption coefficient ?a is 7.17%,and the average relative error of the reduced scattering coefficient ?s' is 5.73%. In the case of adding a certain signal-to-noise ratio (SNR) noise, the iterative inversion method can still obtain higher accuracy of the optical properties.Because of the various restrictions of the optical approximation model, the machine learning method was used to establish the prediction model of the optical parameters. The hyperspectral scattering imaging system based on steady-state spatial resolved technology was used to acquire scattering images of tissue simulating liquid on the wavelength range of 530-900nm. Combining the Fourier decomposition and least squares support vector machine algorithm to establish the nonlinear inversion regression model of optical parameters. The results showed that the modeling method developed by Fourier decomposition and least squares support vector machine based on the experimental data can obtain better predictions.The average relative error of ?a and ?s' is 11.03% and 5.73%, respectively.The prediction model of apple firmness and soluble solids content (SSC) was studied.The on-line hyperspectral scattering imaging system was used to collect hyperspectral scattering images on the wavelength range of 500-1000nm of 'Golden Delicious'(GD),'Jonagold'(JG) and 'Delicious'(RD) apple samples harvested in 2009 and 2010. The optical properties method, moment method, Fourier decomposition method were used to analyze the hyperspectral scattering image and extract the spectral characteristics, and the partial least squares (PLS) and least squares support vector machine (LSSVM) were used to establish the prediction model of apple firmness and SSC. The results showed that compared with the results of the single spectral characteristic, the fused spectral features (optical parameters ?a and ?s' , zeroth-first order moment, Fourier coefficients) can provide more information about the scattering profiles, and improve the prediction accuracy of apple firmness and SSC.