Nondestructive Determination Of Kiwifruit Internal Qualities Based On Hyperspectral Imaging Technology

Fruits occupy an important position in our daily life. With the rapid development of social science and technology, evaluation of the quality of kiwifruit is no longer limited to the external indicators such as color and size. Nowadays, people pay more and more attention to the nutrition and internal quality of fruits. Therefore, it is of great significance that investigate a nondestructive method to determine the soluble solids content(SSC), firmness, moisture content(MC) of kiwifruits to the classification, sales and export of kiwifruit industry.Nondestructive determination of fruit internal quality based on hyperspectral imaging technology has become the focus of many researches, however, using hyperspectral imaging technology to predict the internal quality of kiwifruit is rarely reported. Two varieties of kiwifruits, i.e., “Xuxiang” and “Hayward” were used to predict soluble solids content,firmness and moisture content based on hyperspectral imaging technology in this study.Partial least squares(PLS), least square support vector machine(LSSVM), extreme learning machine(ELM) and error back propagation(BP) models were established to explore the effects of methods of region of interest and areas on the precision of prediction models.Besides, two characteristic variables selection methods, uninformative variable elimination(UVE) and competitive adaptive reweighted sampling(CARS) were compared. The results were as follows:(1) Different areas of hyperspectral image were extracted, the results indicated that the performance of models in the validation set were improved with the areas increasing. The optimal prediction models for soluble solids content, firmness and moisture content were LSSVM modes based on the spectrum extracted by mask. The correlation coefficient(Rp) and root mean square error(RMSEP) of prediction set were 0.8863 and 0.8867°Brix for SSC,0.9403 and 0.7780 for firmness, and 0.8055 and 0.0055 for MC, respectively.(2) The number of characteristic wavelengths selected by UVE for “Xuxiang”,“Hayward” and mixed varieties were 86, 82 and 188 for SSC, 72, 25 and 125 for firmness, 95,60 and 61 for MC, respectively. And the number of characteristic wavelengths selected by CARS for “Xuxiang”, “Hayward” and mixed varieties were 40, 38 and 38 for SSC, 30, 34 and43 for firmness, 30, 33 and 30 for MC, respectively.(3) The optimal model of SSC for “Xuxiang” was CARS-BP, with Rp=0.9384,RMSEP=1.0093 and RPD=2.90, for “Hayward” was CARS-LSSVM, with Rp=0.8452,RMSEP=0.5401, RPD=1.86, and for mixed varieties was CARS-BP, with Rp=0.9438,RMSEP=0.7304,RPD=3.03.(4) The optimal model of firmness for “Xuxiang” was CARS-LSSVM, with Rp=0.9393,RMSEP=0.8361, and RPD=2.91, for “Hayward” was CARS-LSSVM, with Rp=0.8097,RMSEP=0.7650, RPD=1.70, and for mixed varieties was CARS-BP, with Rp=0.8660,RMSEP=0.6900, RPD=1.88.(5) The optimal model of MC for “Xuxiang” was CARS-BP, with Rp=0.6855,RMSEP=0.7158, RPD=1.19, for “Hayward” was FS-LSSVM, with Rp=0.7779,RMSEP=0.6406, RPD=1.58, and for mixed varieties was FS-LSSVM, with Rp=0.8660,RMSEP=0.6900, RPD=1.88.(6) Overall, the results of SSC and firmness prediction for “Xuxiang” were superior to the results for “Hayward”, and the results of MC prediction were inferior to “Hayward”.While the results of SSC and MC prediction for mixed varieties were the best, and for firmness prediction, the results for “Xuxiang” were the best, then were the mixed varieties,the last were “Hayward”.This study indicates that selection of appropriate spectral extraction region is helpful to improve the precision of prediction model. The results demonstrated that hyperspectral imaging technology had potential in nondestructive determination of SSC and firmness of kiwifruit of single variety and mixed varieties, but failed to predict firmness precisely.