Image processing based modeling for Rosa roxburghii fruits mass and volume estimation

Abstract The mass and volume of Rosa roxburghii fruits are essential for fruit grading and consumer selection. Physical characteristics such as dimension, projected area, mass, and volume are interrelated. Image-based mass and volume estimation facilitates the automation of fruit grading, which can replace time-consuming and laborious manual grading. In this study, image processing techniques were used to extract fruit dimensions and projected areas, and univariate (linear, quadratic, exponential, and power) and multivariate regression models were used to estimate the mass and volume of Rosa roxburghii fruits. The results showed that the quadratic model based on the criterion projected area (CPA) estimated the best mass (R2 = 0.981) with an accuracy of 99.27%, and the equation is M = 0.280 + 0.940CPA + 0.071CPA 2. The multivariate regression model based on three projected areas (PA 1, PA 2, and PA 3) estimated the best volume (R2 = 0.898) with an accuracy of 98.24%, and the equation is V = − 8.467 + 0.657PA 1 + 1.294PA 2 + 0.628PA 3. In practical applications, cost savings can be realized by having only one camera position. Therefore, when the required accuracy is low, estimating mass and volume simultaneously from only the dimensional information of the side view or the projected area information of the top view is recommended.