Inversion Of Optical Parameters Of Apple Tissue Based On Hyperspectral Information

Fruits occupy an important position in the daily diet of people.With the improvement of people’s living standards,people’s requirements for fruit quality are getting higher and higher.Although China is a fruit-growing country,the export of fruit is not in the forefront of the world.In addition to its geographical location and climatic factors,high-quality planting and rapid detection of fruits have become an important factor in limiting the fruit exports.The collection and analysis of fruit hyperspectral information has always been a hotspot in non-destructive testing of fruit quality researching.The quality of fruit includes indicators like sugar content,moisture,hardness and so on.These indicators are directly related to the physiological and biochemical status of fruits.The optical parameters directly reflect the physiological and biochemical status of biological tissues.Therefore,optical parameter inversion based on hyperspectral image is an important part of fruit quality testing research.This paper will provide new research ideas for the inversion of apples’optical parameters and even multi-layer biological tissues,study the related techniques and methods for inverting optical parameters of hyperspectral information.The specific research contents are as follows.Parameter measurement:Vernier caliper was used for collected the shape parameters of 200 Shaanxi Red Fuji apples;the hyperspectral information of apple samples was collected by hyperspectral imaging technology;apple peel’s and pupl’s optical parameters of were collected by double integrating sphere system;using sugar meter,electronic balance and the dryer,the apples’ brix content and water content were measured.The simulation data acquisition:1)The surface geometry of the apples were simulated by the surface equation,and the range of the surface equation parameters were determined by the measured apple dimensions.2)The reasons for malformed apples model were analyzed.Based on this,the apple model library construction algorithm was proposed.A Monte Carlo method based on the apple surface equation was proposed to simulate the transmission process of photons in the apple model,and the apple surface brightness simulation images were obtained.Parallelization algorithm was used to improve the operating efficiency of the algorithm.The optical parameter inversion model of simulation data was established.According to the training experience of convolutional neural network,the automatic training and generation algorithm of convolutional neural network was proposed.A convolutional neural network is generated for the simulated data and the optical parameters were classified and inverted.On this basis,the effects of different normalized values,optical parameter combinations and different input positions on the inversion of optical parameters were analyzed.The optical parameter inversion classification effect of the simulated data are:peel absorption coefficient μal:51.55%,pulp absorption coefficient μa2:95.24%,peel scattering coefficient μ1:67.83%,pulp scattering coefficient μs2:90.66%.The deconvolution method was used to verify the source distance of the simulated data.The optimal position of the probe is 1-11 pixels from the incident point of the photon.The optical parameter inversion model of measured spectral data was established.Based on the convolutional neural network of simulated data training,combined with the idea of model migration and feature migration,an MMD-CMS migration method based on maximum mean difference was proposed.Based on the parameter inversion model,an optical parameter inversion model for the measured spectral data was constructed.The optical parameter classification inversion results of the final measured spectral data are:μa1:84.61%,μa2:92.47%,μs1:83.56%,μs2:86.53%,which is the inverse of the inversion model of the convolutional neural network trained directly using the measured spectral data.The performance accuracy increased by 33.06%,17.45%,17.23%,and 20.71%,respectively.Inversion of optical parameters of measured spectral data:The least squares regression method was used to analyze the correlation between the data obtained from the optical parameter inversion model of the measured spectrum and the data of apple’s sugar content and moisture.The correlation coefficients were 0.91 and 0.89,respectively.The correlation coefficient between sugar content and moisture is 0.19 and 0.15 higher than that of direct hyperspectral prediction.The optical coefficient predicts the sugar coefficient and moisture coefficient of the pavilion by 0.17 and 0.14.The results prove the necessity of inversion optical parameters and the effectiveness of inversion model.