Machine vision system combined with multiple regression for damage and quality detection of bananas during storage

Abstract

The study used a machine vision system and applied image process analysis to assess the bruise parameters of the damage zone and external quality attributes of 18 treatments generated from bruised (30 cm and 60 cm drop heights) and non-bruised (control) ‘Fard’ and ‘Somali’ banana cultivars stored at 5, 13, and 22 °C for 21 d. Experiments include digital image analysis of fractal dimension (FD), grayscale (Igray), bruise susceptibility (BS), color, surface area (AS), and some other physical attributes like weight loss %, peel thickness reduction %, and firmness. Multiple linear regression analysis was programmed to determine the effect of the four predictors (storage time, temperature, impact level, and cultivars) on the dependent variables (bruise intensity and quality parameters). It was demonstrated that the fractal dimension of the bruised area increased as impact damage and storage duration increased, particularly in high impact (60 cm) ‘Fard’ banana fruit at 5 and 22 °C, respectively. Digital image analysis was effectively used to determine the surface area of both cultivars. The model explains 83 % of the variability in digital image surface area compared to 70 % for manually measured surface area. The predicted versus measured L*a*b* color spaces showed a higher determination of coefficient (R2>0.5300) compared to a lower determination of coefficient (R2<0.3900) for RGB color spaces. The L*a*b* color spaces recorded a strong correlation with all digitally processed parameters, peel thickness, firmness, and weight loss %, which had greater reliability than RGB color space to predict subsequent bruise damage and storage quality change in bananas. Using image analysis by machine vision system aligning with multiple linear regression has been approved to be efficient for external quality and bruise intensity determination in banana cultivars.