Nondestructive quality evaluation of apples using near infrared reflectance spectroscopy | |
| Author | Jumanazarovich, Chyngyz Erkinbaev |
| Call Number | AIT Thesis no.PH-03-4 |
| Subject(s) | Near infrared spectroscopy Apples |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering |
| Publisher | Asian Institute of Technology |
| Abstract | Multivariate models for non-destructively predicting internal quality of Red Delicious apples such as, total soluble solids (TSS), total acidity (TA) and fruit firmness (FF) were developed from NIR reflectance measurements using a spectral range from 1100 to 2500 nm. To estimate the quality indices of apple samples, the multiple linear regression (MLR), partial least square regression (PLSR), and artificial neural networks (ANNs) models were developed. The three models were compared, based on the correlation coeffcient R and root mean standard error of prediction (RMSEP). The data were devided into calibration and prediction sets. The calibration data set was used to select the wavelengths best correlated with the constituent and to fit a MLR equation and later calculate the constituent value in the prediction data set. The most significant R-values of 0.939, 0.907, 0.723 and corresponding RMSEP of 0.381, 0.0221 and 0.098 were found for TSS, TA and FF of apples, respectively. The overall statistics in prediction by PLSR method showed R-values of 0.961, 0.908 and 0.517 with RMSEP of 0.233, 0.0159 and 0.098 for TSS, TA and FF, respectively. In addition, three-layer backpropagation neural networks (NNs) were developed with different number of neurons in first input layer. The best network models were selected for predicting TSS, TA and FF with R-values of 0.955, 0.872 and 0.67 and RMSEP of 0.269, 0.0185 and 0.087, respectively. Both TSS and TA could be predicted with very good accuracy by all three methods. However, the fruit firmness was not predicted so well. In general, the best results of were obtained from PLSR analysis when discarding the extreme outliers. The MLR technique could also be used based on several selected wavelengths. The artificial NNs approach can be used with selected input data in wavelength regions with more prediction accuracy and without pretreatments of the NIR spectral data. It was concluded that the NIR spectroscopic method seemed reliable for non-destructive and rapid determination of internal quality of apples in terms of TSS, TA and FF with possible application in online grading systems. |
| Year | 2003 |
| Type | Thesis |
| School | School of Environment, Resources, and Development (SERD) |
| Department | Department of Food, Agriculture and Natural Resources (Former title: Department of Food Agriculture, and BioResources (DFAB)) |
| Academic Program/FoS | Postharvest and Food Process Engineering (PH) |
| Chairperson(s) | Jindal, V. K.; |
| Examination Committee(s) | Athapol Noomhorm;Rakshit, S. K.;T. Warunee |
| Scholarship Donor(s) | Asian Development Bank;Japan Government |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2003 |