Nondestructive Detection Of Pear Quality With The Combination Of Dielectric Spectrum And Chemometrics

Fruit is another major economic crop following the grain and vegetables that can bring great economic benefit. Pear is one of China's main fruit, but our country in fruit planting, purchase, transport, storage and processing etc. in the process are still mainly rely on the human labor and experience to evaluate, sort and grad. The lack of effective real-time quality detection device leads to the lower level of the pear postharvest. Therefore, to study the fast and effective quality detection and sorting of pear quality is an important problem to be solved.Based on dielectric properties of nondestructive detection technology and chemometric methods, this study established prediction model of soluble solid content(SSC), firmness of Dangshan pear during ripening, also established prediction model of SSC, firmness and moisture content of postharvest Dangshan and postharvest Korla Fragrant Pear respectively. The mainly chemometric methods used in the study include:Kennard stone (KS) method, random selection method (RS), based on joint X-Y distances method(SPXY), standard normal variety(SNV), uninformative variable elimination (UVE), successive projection algorithm (SPA), partial least squares regression(PLS), back propagation neural network(BP), extreme learning machine(ELM), Generalized regression neural network(GRNN) and least squares support vector machine(LSSVM). Then studied the effect of different combination in forecasting and recognizing model, selecting the optimal combination model. The main conclusions as follows:(1) In the quantitative analysis process, SPXY was the most reasonable, and the model based on SPXY had a great prediction; the derivative algorithm and SNV pretreatment of dielectric spectrum had a little effect on improving prediction results. So we used this full spectrum (FS) data to analysis.(2) For the Dangshan pear during ripening, SPA-ELM model had the best prediction effect of SSC and firmness. For SSC, the Re, Rp, RMSEC and RMSEP were 0.8965,0.8383, 0.4684 and 0.4643 respectively; for firmness, the Rc, Rp, RMSEC and RMSEP were 0.8764, 0.8651,0.4670 and 0.3622 respectively. So the SPA-ELM model could be used to predict the SSC and hardness of Dangshan pear during ripening.(3) For the postharvest Dangshan pear, in SSC prediction models, using the SPXY to divide the samples, the LSSVM model based on FS had the best effect of comprehensive prediction, Rc, Rp, RMSEC and RMSEP were 0.9742,0.9308,0.5923 and 0.8680 respectively, which could be used to detect the SSC of postharvest Dangshan pear; in firmness prediction models, using the SPXY to divide the samples, the ELM model based on UVE had the best effect of comprehensive prediction, Rc, Rp, RMSEC and RMSEP were 0.7816,0.6311, 0.3586 and 0.2855 respectively, but the correlation was still poor, which could not accurately predict the firmness of postharvest Dangshan pear; in moisture content prediction models, using the SPXY to divide the samples, the LSSVM model based on SPA had the best effect of comprehensive prediction, Rc, Rp, RMSEC and RMSEP were 0.9183,0.8732,1.1744 and 1.2213, which could predict the moisture content of postharvest Dangshan pear.(4) For the postharvest Korla Fragrant Pear, in SSC prediction models, using the SPXY to divide the samples, the ELM model based on SPA had the best effect of comprehensive prediction, the Rc and Rp, RMSEC and RMSEP were 0.943,0.876,0.298 and 0.330 respectively, which could be used to detect the SSC of postharvest Korla Fragrant Pear; while the correlation coefficients of firmness prediction models were all less than 0.6, the selected model on the firmness had the poor predictive ability; same as the firmness, we could not found the best prediction model to accurately predict the moisture content, which needed further study.