Skip to main content
ARS Home » Midwest Area » East Lansing, Michigan » Sugarbeet and Bean Research » Research » Publications at this Location » Publication #312673
Title: Automated prediction of sensory scores for color and appearance in canned black beans (Phaseolus vulgaris L.) using a color imaging technique

Author
item MENDOZA, FERNANDO - Michigan State University
item KELLY, JAMES - Michigan State University
item Cichy, Karen

Submitted to: International Journal of Food Properties
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/26/2015
Publication Date: 4/5/2017
Citation: Mendoza, F., Kelly, J., Cichy, K.A. 2017. Automated prediction of sensory scores for color and appearance in canned black beans (Phaseolus vulgaris L.) using a color imaging technique. International Journal of Food Properties. 20(1):83-99.

Interpretive Summary: In this research, a machine vision system was implemented and tested for automatic inspection of color and appearance sensory scores. Commercial canned black beans (Phaseolus vulgaris L.) were evaluated for various color and textural image features extracted from digital color images to characterize and predict the quality of color (COL) and appearance (APP) rates as predicted by a professional visual inspection. Thus, average and standard deviation measurements of color and textural image features (contrast, correlation, energy and homogeneity measurements) from RGB, L*a*b*, HSV color scales were extracted from beans and brine images and evaluated to predict the quality rates of a group of panelists using multivariate statistical models. A total of 69 canned black beans from 18 different brands and 5 markets surrounding Lansing, Michigan, U.S.A., with different production dates (including those labelled as reduced in sodium, no salt added, low calorie, and organic) were collected for this study. Panelist’s rates using 5- quality categories showed that in spite of the ‘fair’ concordance among raters, the information flow deriving from digital color image processing have potential to model, based on partial least squares regression models, the quality scores for APP and COL of a group of bean panelists with an accuracy of 0.937 and 0.871, respectively. Also, a simple classification model, based on support vector machine method, successfully sorted by COL or APP and their combination the quality rates of canned black beans in ‘unacceptable’ and ‘acceptable’ quality groups, even when the panelists’ concordance for sorting beans into two groups was only moderate. Using only three image features for COL as well as for APP average classification performance were 97.2% and 93.9%, respectively. Sorting the beans by COL and APP simultaneously, four image features were needed reaching a classification performance of 89.7%. Overall results showed that statistical modeling of sensorial evaluations for APP and COL using color and texture image data could be used to predict and sort the quality rates of experienced raters of canned black beans in practical industrial application. However, it is necessary to reiterate the importance of the development of suitable quality charts for COL and APP of canned beans, as those proposed here for canned black beans, with the aim of improving the design and development of robust mathematical models for automatic prediction and sorting of bean qualities.

Technical Abstract: BACKGROUND: Evaluation of canning quality of beans is commonly carried out by simple visual inspection by a sensory panel that is in general subjective and limited by the experience of the panelist. The evaluation is further complicated since standard scales to rate visual quality traits of canned beans have not yet been implemented in the trade. In this research, a machine vision system was implemented and tested for automatic inspection of color (COL) and appearance (APP) in canned black beans. Various color and textural image features (average, standard deviation, contrast, correlation, energy and homogeneity measurements from RGB, L*a*b*, and HSV color scales) were extracted from drained/washed beans and brine images, and evaluated to predict the quality rates for COL and APP of a group of bean panelists using multivariate statistical models. A total of 69 commercial canned black bean samples from different brands and markets were acquired and used for analysis. RESULTS: In spite of the ‘fair’ agreement among the sensory panelists for COL and APP, as determined by multirater Kappa analysis (K_free of 0.23 for COL and 0.21 for APP), a machine vision data based on partial least squares regression (PLSR) analysis showed high predictive performance for both COL and APP with correlation coefficient for prediction (R_pred) of 0.937 and 0.871, standard error of prediction (SEP) of 0.26 and 0.38, and the ratio standard deviation to SEP (or RPD) of 2.8 and 2.0, respectively. When a classification was performed based on both COL and APP traits, a support vector machine model using simple image data was able to sort the canned bean samples into two sensory quality categories of ‘acceptable’ and ‘unacceptable’ with an accuracy of 89.7%. CONCLUSION: Using simple color and texture image data, a machine vision system showed potential for the automatic evaluation of canned black beans by COL or/and APP as a professional visual inspection.