Nondestructive Detection Of Fruit Internal Quality Based On Portable Near Infrared Spectrometer

China has very rich fruit resources and is the biggest fruit producer, but the low level of postharvesting technologies (handling, processing, grading and sorting) leads to the poor quality and lack of international competitive power in the world market. The traditional methods for fruits'quality detection are low-tech, slow and easily disturbed by man-made factors, while near infrared (NIR) spectroscopy can achieve non-destructive and fast quality detections.In this paper, Detection of apple soluble solids content (SSC), totle acidity and firmness of apple were studied using near infrared spectroscopy with chemometrics analysis, and quantitative models were established on NIR spectra for apple SSC, TA, and firmness.In the research studied using a commercial portable near infrared spectrometer (K-BA100R), for determination of SSC and TA stepwise multi linear regression(SMLR) and least square support vector machines(LS-SVM) basing on selected wavelengths, and partial least square regression(PLSR) based on full spectral range were compared. The result showed that the performance of SMLR models established on raw spectra were better than LS-SVM and PLSR. For SSC prediction, SMLR model based on 6 selected wavelengths gave the highest correlation coefficient (r) to 0.980, the lowest root mean square error of prediction (RMSEP) to 0.238°Brix. For TA prediction, SMLR model based on 11 selected wavelengths gave the highest r to 0.866, the lowest RMSEP to 0.115.For quantitative detection of apple firmness, several pretreating methods were used to improve the PLS models and compared. The result showed that PLS models based on raw spectra gave better performance than the others, with r and root mean square error of calibration (RMSEC) of 0.782 and 1.98N/cm2 for calibration set and r and mean square error of prediction (RMSEP) of 0.613 and 3.12N/cm2 for prediction set, respectively. In the research studied using a simple VIS-NIR instrument equipped with USB 4000 spetrometer. First the portable VIS-NIR equipment was briefly introduced, then different PLS models based on different pretreated spectra were compared and studied for apple's SSC and TA prediction, the result showed that PLS model based on spectra pretreated in SNV method gave better prediction for SSC,and the MSC method used to raw spectra could improve the performance of PLS model in TA prediction; successive projections algorithm (SPA) method was used to select wavelengths to establish MLR model to predict SSC, and the result was better than PLSR models. Finally, some suggestions were made to improve the equipment.