Grading method for tomato multi-view shape using machine vision

Owing to the requirements of a high yield and high-quality tomatoes, tomato grading is important-particularly for fruit morphology, and accuracy has become the focus of attention. Machine vision provides a fast and nondestructive manner to address this demand. In this study, the gamma correction method was used for preprocessing to enhance the edge information of tomatoes, and Otsu's method was used to eliminate the tomato-image background in the A-component image under the LAB color model. On this basis, two levels of exploration were conducted. First, new evaluation indices were proposed for tomato shapes from different views. For the top view, two shape-evaluation indices were established: the area ratio of the maximum inscribed circle to the maximum circumscribed circle and the dispersion of the contour centroid distance (range and coefficient of variation), the highest accuracy was 94%. For the side view, the difference between the maximum and minimum centroid distances in the contour was established as a shape index, the highest accuracy was 91.91%. Second, an evaluation method based on multi-view fusion was developed by combining the advantage indices for different views. The classification accuracy reached 96%, with the highest identification accuracy of unqualified tomatoes. The results show that the proposed evaluation method combining top views (dispersion of centroid distance) with side views (difference between maximum and minimum centroid distances) is effective for tomatoes.