On-line Monitoring Of Moisture Ratio For Fruits And Vegetables During Vacuum Freeze-drying Based On Image Processing Technique

Vacuum freeze-drying (VFD) technology, which is a very popular and advanced method which could maintain original flavor, nutrients color and shape quite well for fruits and vegetables, has a wide application prospect for processing of agricultural products and food. However, high energy consumption and lower profit restrict the development and application of VFD technology. Therefore, optimization of process parameters has received wide attention by many researchers in recent years.Color meament and texture analysis technique, which were included in computer vision and image processing, were competent to monitor moisture rate on-line for fruits and vegetables during vacuum freeze-drying process. Color values and texture values were extracted and regressive models were produced between moisture rate and the values. Besides, water boundary was captured and extracted in the cross-section image of material and its movements were computed by micro-strain field, a model between moisture rate and displacement of water boundary was established at the same time. The main research contents of this subject include:(1) Apple、potato、eggplant carrot and banana, which were familiar with people, were chosen as experiment material, the five samples were divided into fifteen blocks according to the following three treatments:original samples; dyed samples; high pulsed electric field pretreatment samples. A CCD digital camera with polarizer was fixed on a tripod and placed in front of freeze-drying container. The digital camera could acquire all serial images of sample surface during vacuum freeze-drying processing period. Images consisted of1024×768pixels were saved in.jpeg mode at1.875fps.(2) Color features were extracted in RGB and L*a*b*color space, and a model was established between R、G、B、L、a*、b*color values and moisture rate of samples. It was found that a*and b*values were not dependent on the time, so there were no a*and b*color values in the fifteen models. Statistical analysis concluded that there were higher determination coefficient (R2>0.85) and p values of the models were <0.0001. As far as regressive coefficients, there were the same conclusion as models. Furthermore, most absolute errors were controlled0.01%～5%and relative errors of models were about5%based on gravimetric method (GB5009.2-85). Therefore, the fifteen models are feasible and this method was a good way to estimate moisture ratio of fruits and vegetables.(3) Established the moisture rate models using principal component analysis, and the first principal component (Prinl) instead R、G、B and L color values. Although cumulative proportion of Prinl were all>0.90, determination coefficient of the models between Prinl and moisture rate were very different.Therefore, it was thought that using principal component analysis based on color measurement method alone had poor adaptability.(4) Mean value, standard deviation, smoothness, the third moment, uniformity and entropy were extracted and analyzed statistically, fifteen models were established between these texture feature values and moisture rate. As far as the models, Prinl which could represent the6texture feature values was computed by principal component analysis method, the cumulative proportion of Prinl are almost>0.80, the maximum was0.9494, and minimum was>0.70. Prinl and Prin2should be used to represent moisture rate synchronously, and the main factor was Prinl, the others were observation error.Models were established between Prinl and moisture rate. The results indicate that all the p values of models were0.0001, whose coefficient of determination were>0.80. In addition, all the p values of regressive coefficients and intercept were0.0001, therefore, the fifteen models could predict moisture rate feasiblely according to the texture feature analysis method.(5) Using color feature values and texture feature values jointly to predict moisture rate by regression analysis. The results showed that determination coefficient of the fifteen models were0.9999～1.0000, all the p values of models were0.0001, model checking very significant and high fitting accuracy. This showed that take color feature values and texture feature values into account would make moisture rate models more accurate and more practical.(6) Using displacement field detection methods, materials internal moisture diffusion and migration law during vacuum freeze-drying process were analyzed quantitatively through micro-displacement amount. The results showed Otsu method, K-Means method, Pseudo-color image processing method, and Sobel edge detection method were satisfactory method which could extract boundry. Material geometric center as the origin of the coordinate system is established, the Harris corner detection method to find the corner points on the water boundry, and extract coordinate values of corner points which intersect the coordinate axises, displacement of the points were calculated by1h interval. Correlation analysis showed that the p values of models was <0.0001, whose coefficient of determination was0.9998, model checking very significant and high fitting accuracy. Test results of regressive coefficient showed that the micro-displacement of the four corner points and moisture-rate-squared of the material were very significant. In other words, moisture rate could be expressed and predicted by the displacement field of the water boundary parameters (the amount of micro-displacement).A new on-line moisture monitoring method was proposed about process controlling and optimization of process parameter during vacuum freeze-drying process. In summary, the mentioned models and methods not only provide a new monitoring method of moisture ratio, but also give foundation of monitoring moisture ratio for other drying processes.