Image-based parameters: Bayesian Model in banana fruit mass prediction

Abstract

The emergence of contagious diseases has intensified the need for the adoption of technologies that minimize human contact with
potentially contaminated surfaces and objects. Due to the manual nature of fruit weighing within Zimbabwean supermarkets, there is
frequent human-mass scale contact. Hence, manually operated mass scales are on the long list of prospective disease spreading
surfaces. This study proposes the assessment of the feasibility of banana fruit mass modelling based on image analysis, as a way of
eliminating human-mass scale contact during the manual weighing process. The banana fruit image-based parameters considered in
this study are filled image, minor axis length, equivalent diameter, perimeter, stalk length and major axis length. The Bayes Linear
Regression model has been utilized in identifying image-based parameters that are of significance to the mass determination model.
The major axis length of the banana has been identified as the most significant mass prediction image-based parameter. Using the
Mean Absolute Percentage Error (MAPE), the major axis length-based model accuracy has been assessed. The model has an
accuracy of 96.61 %. Since, the accuracy value lies within the upper quartile region, the formulated model is accurate enough to be
used for banana mass prediction. Based on the paired t-test results, the difference between the average of predicted value minus
actual value is not big enough to be statistically significant. The system has an acceptable response time of four seconds.