PREDICTING SENSORY RESPONSES OF BANANAS BY SMARTPHONE IMAGE BASED MODELS

The color of the banana peel is an indicative of the fruit’s level of ripeness, which influences sensory characteristics of taste and texture. Thus, this study aimed to evaluate the potential of using smartphone images to predict consumer sensory responses to bananas. Sensory and color data were collected from 85 banana samples, being analyzed 7 different ripening stages (from green to yellow with brown edges). Color data were obtained two ways: (i) through a colorimeter (parameters L*, a*, b*, c*, and h°) on 6 different points; and (ii) through smartphone photographs (parameters R, G, and B). Sensory analysis were performed to assess the overall acceptance, the ideal sweetness, and the ideal firmness, being collected 90 sensory responses on each stage of maturation. Multiple Linear Regression (MLR) models were used to predict sensory responses employing the color parameters. A validation set, external to the calibration set, comprised 25% of the observations. Models based on the colorimeter data had a calibration coefficient R² >0.95 and a validation coefficient R² >0.87 for predicting ideal sweetness and firmness, with Root Mean Square Error (RMSE) <0.09 and 0.16 for calibration and validation, respectively. However, for global acceptance, the model presented calibration coefficient R² >0 .94, but a validation coefficient R² <0.51. On the other hand, photograph-based models had calibration coefficient R² >0.92 and validation coefficient R² >0.91 in predicting ideal sweetness and firmness, with RMSE <0.21 and 0.31 for calibration and validation, respectively. Regarding the global acceptance, this model presented calibration R² >0.87 and >0.96 for validation. Overall, smartphone images performed better as descriptors for predicting sensory responses than using colorimeter data, which may be due to the use of colorimetric data from the entire banana when using digital photographs.