Tomato weight prediction based on image processing

Abstract

Objective： A tomato weight detection method based on image processing was established to realize non-contact tomato weight detection. Methods： The binary image of tomato was obtained through image processing. The geometric features of tomato were extracted by pixel statistics and minimum circumscribed rectangle method, and correlation analysis was made between the characteristics and the real value of tomato weight, then the regression model of tomato weight detection with geometric features as parameters was established. Results： Compared with the real size of tomato, the measurement error of transverse and longitudinal diameter of Tomato by minimum external rectangle method was less than 3%. In addition to fruit shape index, other geometric characteristics were linearly correlated with tomato fruit weight, and the correlation between positive characteristics and fruit weight was more significant. Three types of 20 models were established for prediction and evaluation. The multiple regression model with the parameters of tomato front projection area and perimeter, projection area of a side image and tomato transverse diameter had the highest accuracy, the regression coefficient was 0.962, the average relative error of detection value was 0.673%, and the average absolute error was 1.425 g. Conclusion： The model is suitable for the weight detection of tomatoes and other fruits or articles with similar axisymmetric shape characteristics.

Keywords

• machine vision
• image processing
• feature extraction
• tomato weight