Rapid Detection Of Vegetable Freshness Based On Hyperspectral Imaging Technology

Freshness is the most important indicator for the quality of leafy vegetables. At the same time, the quality of vegetables has gradually become the basic criterion which determines whether they can establish a market domestic and overseas. The precise production and management of vegetables count on advanced technology of production, storage, evaluation and classification based on freshness. So it’s urgent to develop researches about the selection of vegetable freshness indexes and evaluation methods aimed at ensuring nutrition, security and cleanness of the vegetables people consume. Thus contribute to substantial quality improvement of vegetables and reasonable utilization of vegetable resources. Spinach’s richness in vitamins, protein, minerals and dietary fiber makes it a nutritious and delicious aliment. However, postharvest spinach has high metabolism and respiratory rates, which result in great tendency to wilting and decay.In this study, we used "universal spinach" which is common on the market as the research object, extracted the hyperspectral imaging information and at the same time measured the content of Chlorophyll a, Chlorophyll b, total Chlorophyll, Carotenoid and soluble sugar in spinach leaves with laboratory chemical measurements. Then used chemometrics methods to establish freshness discriminant models and above-mentioned chemical indexes prediction models based on the spectral information, respectively. The results turned out good, and the main conclusions are as follows:(1)Based on visible/near infrared hyperspectral information to discriminate spinach freshness, samples stored at4℃and20℃were used to establish full spectrum PLS-DA and SR-PLS-DA, SR-SVM and SR-ELM discriminant models, respectively. Among them, the SR-ELM models of4℃and20℃both achieved the best prediction results, with recognition rates reaching to100%. The effect of SR-PLS-DA took the second place. For samples of4℃,the recognition rate of test set was97.50%and was higher than that of full spectrum PLS-DA model(95.00%). For samples of20℃, the recognition rate of prediction set was100%and was much higher than that of full spectrum PLS-DA model(86.70%).The results of SR-SVM models were the worst for both samples of4℃and20℃, with the recognition rates of prediction sets being around70%.(2)Based on near infrared hyperspectral information to discriminate spinach freshness, samples stored at4℃and20℃were used to establish full spectrum PLS-DA and VIP-PLS-DA, VIP-SVM and VTP-ELM discriminant models, respectively. Among them, the VIP-ELM models of4℃and20℃both achieved the best prediction results, with recognition rates reaching to100%. The effect of VIP-PLS-DA took the second place. For samples of4℃,the recognition rate of prediction set was92.50%and was lower than that of full spectrum PLS-DA model(100%). For sample of20℃, the recognition rate of prediction set was100%and was as high as that of full spectrum PLS-DA model(100%).The results of VIP-SVM models were the worst for both samples of4℃and20℃, with recognition rates of prediction sets being lower than70%.(3)The rapid chemical contents detection of freshness indexes were conducted on samples of4℃.The best prediction model for Chlorophyll a was SPA-ELM model and the prediction result of test set was:rp=0.8062, RMSEP=0.2697; The best prediction model for Chlorophyll b was SPA-MLR model and the prediction result of test set was:rp=0.6844, RMSEP=0.0819; The best prediction model for total Chlorophyll was SPA-ELM model and the prediction result of test set was:rp=0.8500, RMSEP=0.2719; The best prediction model for Carotenoid was SPA-BPNN model and the prediction result of test set was:rp=0.7656, RMSEP=0.0587; The best prediction model for soluble sugar was SPA-MLR model and the prediction result of test set was:rp=0.8499, RMSEP=0.9104. Above results showed that it was feasible to rapidly detect contents of Chlorophyll a, Chlorophyll b, total Chlorophyll, Carotenoid and soluble sugar in spinach leaves stored at4℃based on visible/near infrared hyperspectral imaging technology, and the prediction results showed a relatively satisfactory accuracy. For samples stored at4℃, total Chlorophyll achieved the best prediction result and Chlorophyll b did the worst.(4)The rapid chemical contents detection of freshness indexes were conducted on samples stored at20℃.The best prediction model for Chlorophyll a was SPA-ELM model and the prediction result of test set was:rp=0.9255, RMSEP=0.1991; The best prediction model for Chlorophyll b was SPA-MLR model and the prediction result of test set was:^=0.8560, RMSEP=0.0577; The best prediction model for total Chlorophyll was SPA-ELM model and the prediction result of test set was:rp=0.8792, RMSEP=0.1739; The best prediction model for Carotenoid was SPA-MLR model and the prediction result of test set was:rp=0.8929, RMSEP=0.0274; The best prediction model for soluble sugar was SPA-ELM model and the prediction result of test set was:rp=0.8719, RMSEP=0.7575. Above results showed that it’s feasible to rapidly detect contents of Chlorophyll a, Chlorophyll b, total Chlorophyll, Carotenoid and soluble sugar in spinach leaves stored at20℃based on visible/near infrared hyperspectral imaging technology, and the prediction results show a relatively satisfactory accuracy. For samples at20℃, Chlorophyll a achieved the best prediction results,Carotenoid took the second place and Chlorophyll b did the worst.(5)The prediction results of five indicators for samples stored at20℃were better than those for samples at4℃.Both of these two sample groups worked worst for the prediction of Chlorophyll b content, while other four indicates showed relatively good prediction results. Chlorophyll a, total Chlorophyll, Carotenoid and soluble sugar could be used as spinach freshness indicators for further research.