Volume Prediction Method for Pomelo Based on Symmetry Axial

Highlights Multi-angle based approach for pomelo dataset collection improves the robustness of volume estimation models. Symmetry axis feature extraction is more suitable for volume estimation of pomelo fruit shape. BP Neural Network-based estimation model shows better volume estimation ability. Abstract. Pomelo quality grading helps to control the quality of pomelo production and promotes a standardized production. The realization of automated pomelo quality grading relies on the acquisition of several phenotypic parameters, with pomelo volume being a crucial one. Fruit phenotypic features, such as projected area, perimeter, radial distance, and the maximum radius of the tangent circle, are often used to estimate fruit volume through image processing. To enhance the accuracy of volume estimation, this article introduces a pomelo volume estimation model that relies on image features. This model is based on an analysis of the fruits’ symmetry axis and combines various features, including the projected area, perimeter, radial distance, and the maximum radius of the tangent circle. In this study, we collected a dataset of 3,200 pomelo images and compared the prediction results under different parameter combinations. After incorporating the parameters associated with the symmetry axis, the root mean square error of volume prediction was 159.82. The coefficient of determination was 0.84, with a mean absolute error of 142.67 and a mean absolute percentage error of 0.23. Notably, the coefficient of determination was improved by 0.26 compared to the results obtained without considering the symmetry axis features. The prediction error falls within the acceptable range, meeting the requirements of practical application. Overall, the proposed method provides a valuable technical reference for pomelo quality grading. Keywords: Axis of symmetry, Feature parameters, Machine learning, Volume prediction.