| It is the hot research project of fruit quality in-line measurement and classification in international agricultural engineering field, which will satisfy the consumer's demand in fruit quality and safety, improve the products value and competition and increase the farmers' income. Watermelon is one of people's favorite fruits and widely grown in more than 90 countries on five continents. The yield of watermelon in China exceeds 35000 thousand tons annually and ranks first in the world. However, the export proportion in trade is very little and most of them used for domestic consumption. There is a potential in the export trade. An important factor affecting the export trade is that there is no reliable means to guarantee the internal quality. Furthermore, with the enhancing of the peoples' standard of living, consumers pay more and more attention to its internal quality, such as taste, sweetness, and other ingredients. Soluble solids content is one of main quality indicators, which is related to sweetness. In addition, crispness related to firmness, density related to creases or/and voids are also concerned by consumers. At present, the common nondestructive method available is the inspection by slapping or rapping it. The method is subjective and the mistake is inevitable. Therefore, it is urgent to adopt a method feasible in technique and appropriate in economy to detect the watermelon's quality nondestructively. According to the situation mentioned above, the visible and near-infrared (Vis/NIR) spectroscopy method was studied to evaluate the internal quality of watermelon. The main contents and conclusions were as follows:1. The common methods used for evaluation internal quality of watermelon were reviewed, such as density method, vibration and acoustic method and magnetic resonance imaging method, and the limitations of these methods in technique actualization and applicability in economic were analyzed.2. The common Vis/NIR spectra detecting methods and their application fields that used for fruits quality measurements were reviewed. According to the properties including the large difference in sweetness and structure between the central and the outside fruit flesh and the diverse distribution of the properties individually, large-size and thick-skin, the Vis/NIR diffuse transmittance technique was selected here for the internal quality measurement of watermelon, furthermore, the light system and spectra detecting system used in this paper was constructed. 3. According to the large-size and the distinct difference in size, the size effect on the spectra and the results of calibration and prediction was investigated. To reduce or remove size effect, the size classification (volume) elimination method, volume average method and normalization method based on volume were presented and the calibration and prediction results indicate that these three methods could reduce or remove the size effect effectively. The normalization method is more suit to in-line application and obtained better calibration and prediction results with the r of 0.789, 0.858 and 0.947, the RMSEC of 0.596, 0.499 and 0.313, and the RMSEP of 0.606, 0.686 and 0.413 than the results without normalization with the r of 0.716, 0.793 and 0.848, the RMSEC of 0.678, 0.592 and 0.515 and the RMSEP of 0.825, 0.764 and 0.714 for the original spectra, the first derivative spectra and the second derivative spectra with soluble solids content using partial least square method , respectively.4. Owing to much thick skin of watermelon, contrast experiment was conducted in skin-peeled watermelon and intact watermelon to compare the influence of skin on spectra. Results showed that the influence mainly concentrated in visible waveband (650-750nm). Then the waveband was divided into two parts to develop calibration model using partial least square method for soluble solids content to remove the effect of the skin and the results indicated that the measurement results was not be enhanced. However, the difference of calibration results in the wavelength of 650-950nm with the r of 0.949, the RMSEC of 0.292 for intact watermelon and with the r of 0.973 and the RMSEC of 0.214 for skin-peeled watermelon indicated that the influence of skin is not very serious and the intact watermelon spectra also can get satisfied calibration result.5. Vis/NIR spectroscopy nondestructive detection needs high-energy light source, so differences of diffuse transmittance spectra with different light intensity (600W, 1000W) were detected and effects of light intensity on spectra and calibration results were compared. From the results, it can be seen that higher light intensity is helpful to spectra smoothing and improving calibration result, but it should guarantee to have no damage on the skin.6. According to the characteristic that the watermelon stem-calyx grows first and then weights in landscape orientation, we selected calyx, stem and equator as detecting positions for two breeds of watermelons (Qilin and Zaochunhongyu) to evaluate effects of spectra detected from different positions on calibration results in order to choose ideal spectra detecting position. The results show that the spectra detected from the three positions all can get satisfied calibration models and no rule indicates which position can get better spectra for calibration.7. Density and firmness are two important physical indexes of watermelon. Crispness is related to firmness and creases or/and voids in tissue are related to density. Potential of the two indexes being measuredd by Vis/NIR diffuse transmittance technique was studied in experiments. The second derivative spectra calibration and prediction results with the r of 0.910, the RMSEC of 0.015, and the RMSEP of 0.018 for density and the r of 0.916, the RMSEC of 0.308, and the RMSEP of 0.515 for firmness with normalization using partial least square method indicated the feasibility of the two indexes being measuredd by Vis/NIR diffuse transmittance technique.8. According to the fact that farmers always picked unripe watermelon and treated them with high concentration ethylene to quicken ripeness, classification experiments on the two kinds of watermelon mentioned above were conducted based on the Vis/NIR spectroscopy diffuse transmittance technique. In discriminant analysis, a method which classified them by diffuse transmittance ration at two wavelengths was adopted to discriminate them. Result with mistake ratio 32.5% for samples without ethylene treatment and 20% for ethylene treatment samples indicated that this method could discriminate the two classes of watermelons roughly. Mahalanobis distance and partial least square methods were also used here for discriminant analysis and satisfied results were obtained. The first derivative spectra with Norris derivative filtering of samples without being ethylene treated using Mahalanobis distance discriminant analysis got the result with mistake ratio 1.67% for calibration set and no mistake for prediction set and no mistake for samples being ethylene treated. No mistaken took place for the second derivative spectra using partial least square method.9. Several methods, such as classical least square, multiple linear regress method, principal component regression method and partial least square method that always used for developing calibration models were selected to develop the models for watermelon quality index. The results indicated that the partial least square is the most appropriate method for watermelon.10. To make Vis/NIR diffuse transmittance technique apply to watermelon in motion, dynamic spectra detecting system exploited by our laboratory was rebuilt. Moreover, static and dynamic spectra detecting experments were conducted and effects of noises caused by motion on spectra, calibration and prediction results were analyzed and compared. Furthermore, the least-square filtering method, Norris differential filtering method and small wave transformation filtering method were adopted to eliminate effects of noise on spectra smoothing, then the spectra used for calibration and prediction respectrively. Results indicated that Norris differential method was an effective method to smooth spectra and it improved calibration and prediction results. The calibration models were developed for soluble solids content and density with r of 0.865, 0.918, RMSEC of 0.536, 0.016 and RMSEP of 1.030, 0.021 for calibration and prediction results, respectively.11. The fruits internal quality nondestructive on-line measurement experiment platform including Vis/NIR spectra on-line detecting software and dynamic detecting system was established. According to Vis/NIR spectra detecting principle, simple spectral pretreatment and calibration methods, the on-line experiment on watermelon's soluble solid content measurement was conducteded. Results showed that the main function of experiment platform satisfied requirements of on-line internal quality measurement of watermelon real-time. So, parts of functions of the system need further improvements in future research.
|
PDF Full Text Request