Evaluation Of Fruit Quality And Wood Moisture Content Based On Hyperspectral Imaging

Nondestructive detection of fruit quality includes the identification of surface defects and the prediction of internal quality.Bruise is the primary reason why fruits are classified as inferior.The sugar content of pulp has an important influence on the hardness and maturity of fruit.Wood is widely used in papermaking,chemical manufacturing,furniture manufacturing and other fields.Wood moisture content is an important index of wood properties.In this study,hyperspectral imaging system and machine learning method were used to detect the slight damage and sugar content of fruit as well as the moisture content during wood drying.The main contributions are listed as follows:1.A nondestructive quantitative method combining a genetic algorithm modified by ReliefF(ReGA)with support vector machine(SVR)was proposed for predicting sugar content of Korla pear.Hyperspectral images with a spectral range of 400~1000 nm of pears were acquired by hyperspectral imaging system.Then the region of interest(ROI)function of ENVI 5.3 software was used to conduct spectral data extraction from each hyperspectral image of pear and smoothed by the standard normal variate(SNV)method.Totally,157 pear samples were divided into calibration set(105)and prediction set(52)based on the Kennard-Stone(KS)sample set partitioning method.ReGA was applied to extract the characteristic variables from full spectra(FS).The selected characteristic variables and full spectra were taken as the input vector to establish support vector regression(SVR)models respectively.Three wavelengths selection methods including successive projection algorithm(SPA),competitive adaptive reweighted sampling(CARS)and ReliefF were used for comparison.The experimental result indicated that ReGA is an effective wavelength selection method,which can not only simplify the model,but also predict the sugar content of Korla pear accurately.2.Slight damage identification method of Fuji apple based on minimum redundancy maximum relevance(mRMR)and kernel extreme learning machine(KELM)was proposed.Hyperspectral images of both sound and damaged apples at five stages(1 min,1 day,2 days,3 days and 4 days after bruising)were acquired by the SOC710-VP hyperspectral imaging system.Then the average spectral reflection of the appropriate region was extracted for data analysis.For reducing data redundancy and computation overhead,mRMR was applied to selected characteristic wavelengths.After that,KELM and ELM were used to build models to identify the bruised apples.Unsupervised feature selection with ordinal locality(ufsol),Lasso algorithm and Infinite Latent Feature Selection(ILFS)were used for comparison.The results showed that the mRMR-KELM models exhibited the best classification capability,and with the prolongation of damaging time,the classification accuracy increases gradually.3.Based on linear discriminant analysis(LDA)dimension reduction algorithm and partial least squares(PLS),a nondestructive method to predict moisture content of beech was proposed.In the process of wood drying,hyperspectral imaging system is used to collect hyperspectral images of different drying stages.Then the spectral reflection data of each pixel is extracted.To reduce the redundancy between high-dimensional data,using LDA algorithm to reduce the dimension of hyperspectral data,and then establish PLS models to predict the water content.Principal component analysis(PCA)and factor analysis(FA)were also used to reduce the dimension of the original data.The results demonstrated that after the dimension of input data was reduced from 128 to 12 by LDA Algorithm,the results of PLS model was the best.The prediction results of LDA-PLS model are represented by images,which can not only obtain the average moisture content of wood,but also directly observe the distribution of water content,providing reference information for parameter adjustment in the process of wood drying.