Nondestructive Identification Of Kiwifruits Treated With Expansion Agent By Near-infrared Spectroscopy And Hyperspectral Imaging Techniques

As living standards and consumption levels improved, peoples pay more and more attention to the quality of fruit. Kiwifruit dipped in with expansion agent has become a common phenomenon at the beginning of growth. Although expansion agent can increase single kiwifruit weight and total kiwifruit production by 30%~50%, the quality of kiwifruits treated with expansion agent is poorer and the storage period of kiwifruits treated with expansion agent is shorter than that of normal ones. When concentration of expansion agent used is too high, it will severely degrades kiwifruit quality and influences tissue structure and fruit shape. Moreover, the potential effect of expansion agent on people health is not clear. Therefore, developing a nondestructive method for identifying kiwifruits treated with expansion agent is a key problem to be solved in kiwifruit postharvest processing.Near-infrared(NIR) spectroscopy and hyperspectral imaging of kiwifruits picked in 2013 were obtained for this paper. Standard normal variate transformation was used to preprocess original spectra. The samples were divided into calibration and prediction set based on Kennard-Stone method. The modeling methods, such as partial least squares(PLS) and support vector machine(SVM) were applied to establish models for identifying expanded kiwifruits and normal ones based on principal components extracted by principal component analysis(PCA), characteristic wavelengths selected by uninformative variable elimination(UVE), characteristic wavelengths selected by successive projections algorithm(SPA) and full spectra(FS), respectively. Another batch of samples which picked in 2014 were used to test the reliability of models and research the method of improving correct identification rate of models. The conclusions were as follows:(1) In the process of using NIR spectroscopy for the identification of kiwifruits treated with expansion agent, the number of principal components extracted by PCA and characteristic wavelengths selected by SPA was only 0.53% and 0.29% of FS. The number of characteristic wavelengths extracted by UVE was 1228, which had too much input variables. Average correct identification rate of all the models for identifying expanded kiwifruits and normal ones were higher than 96.7% for both calibration and prediction set. Correct identification rate of PCA-PLS model was the highest, whose accuracy rate reached 100.0% for both calibration and prediction set.(2) In the process of using hyperspectral for the identification of kiwifruits treated with expansion agent, the number of principal components extracted by PCA and characteristic wavelengths selected by SPA was only 2.34% and 3.12% of FS. The number of characteristic wavelengths extracted by UVE was 108, which had too much input variables. From the point of view of modeling method, correct identification rate of the models built by PLS was all higher than those of SVM. Correct identification rate of SPA-PLS model was the highest, whose accuracy rate reached 100.0% for calibration set and which was 98.4% for prediction set.(3) In the process of the models based on samples of 2013 tested by the samples of 2014, average correct identification rate of the models built by PLS based on NIR spectroscopy for prediction set was 84.4%, which was 83.3% for SVM. Average correct identification rate of the models built by PLS based on hyperspectral for prediction set was 83.3%, which was 82.2% for SVM. In the process of modifying models by adding new samples from the kiwifruits obtained in 2014, average correct identification rate of the models built by PLS and SVM for calibration set was all higher than 93.0%. For prediction set, average correct identification rate of the models built by PLS and SVM for expanded kiwifruits was higher than 86.0%, which was higher than 83.0% for normal kiwifruits.The results clearly indicated that near-infrared diffused spectra and hyperspectral imaging technique had the potential as an efficient, accuracy and non-invasive method for distinguishing kiwifruits treated with expansion agents from normal kiwifruits. Identification effect of model built by the samples of last year performed pool in validation of the sample picked the next year. Identification rate of the models could be improved by model modification, while there was still room for improvement.