Application Of Chemometrics Combined With Color Space To Classify And Detect Red Bayberry After Postharvest And Processing

Red bayberry (Myrica rubra Sieb.&Zucc), which belongs to the Myricaceae. is a kind of fruit with high medicinal and edible value. It contains high nutritional components such as anthocyanins, sugars, organic acids, phenols and vitamins, and is popular with local people. China is the original and major commercial production area for bayberry. And with the development of preservation and processing technologies, the bayberry industry is also developed. Therefore, it is important to pay attention to the red bayberry postharvest handling in order to increase the income of farmers and the economic and social benefits. One of the most important postharvest handling steps is sorting which always be a hot area of research. Meanwhile, due to without epicarp protection and ripening in the hot and rainy season of June-July, bayberry is highly susceptible to mechanical injury and microbiological decay which will limit the postharvest self life to a very short time. To extend their consumption time, red bayberry is usually processed into various products. However, during processing and storage, it will be undergo a number of deteriorative reactions, resulting in product quality degradation. Therefore, improving the quality detection technology of red bayberry and its processed products is critical.This study was proposed to solve the above issues, and its objectives were (1) to develop partial least-squares support vector machine (LS-SVM) classifiers for detection of bruises on red bayberry as a function of sharp (fractal values) and color feature (nRGB, CIELAB, YCbCr and HSI color spaces),(2) to develop partial least squares regression (PLSR) and least-squares support vector machine (LS-SVM) models for prediction soluble solids content (SSC) and pH in red bayberry based on color spaces (nRGB, CIELAB, CMY, HSI, I1I2I3and YCbCr),(3) to develop support vector machine (SVM) based on color spaces (nRGB, CIELAB, YCbCr and HSI) for evaluating the changes of nutrition in red bayberry juice during storage.After studying, the results can be listed as follows:(1) Our results showed that compared with the other color spaces the LS-SVM classifier based on HSI color space (γ=34.21. σ2-554.35) can be achieved the best classification accuracy which is94%. However, the classification accuracy was much higher by LS-SVM models based on combining fractal values with color spaces. Among them, the highest classification accuracy (98%) was found when using LS-SVM model based on combining fractal values with HSI color space (y=0.65, a2=22.53).(2) The results showed that PLSR and LS-SVM models coupled with color spaces could predict pH value in red bayberry (r=0.93-0.96:RMSE=0.09-0.12; MAE=0.07-0.09; and MRE-0.04-0.06). For predicting SSC, PLSR models based on CIELAB color space (r=0.90, RMSE=0.91, MAE-0.69and MRE=0.12) and HSI color space (r=0.89; RMSE=0.95°rix, MAE=0.73and MRE=0.13) were recommended.(3) In this study, three different parameters optimization algorithm (grid-search algorithm, genetic algorithm and particle swarm algorithm) for SVM were investigated. And the models using grid-search algorithm and genetic algorithm are better than using particle swarm algorithm. The results indicated that all the models showed poor performances for prediction of total phenols and total flavonoids. However, SVM modes based on HSI color space with grid-search algorithm parameters optimization technique provided the best prediction of the changes of anthocyanins (R2=0.910, MSE=0.012mg/100mL) and ascorbic acid (R2=0.930, MSE=0.014mg/100mL) in red bayberry juice during storage.In conclusion, color space combined with chemometric techniques was successfully developed for bruises and quality detection on red bayberry. It not only provides a potential tool for detecting the quality of red bayberry, but also provides some theoretical and practical guidance for the development of food computer vision technique. In addition, according to our study, the chemometric models coupled with HSI color space which is developed based on the concept of visual perception in human eyes are recommended in red bayberry quality detection.