Apple Size Estimation Method with 3D Projection Correction for In-field Grading System

Apples postharvest sorting is crucial for increasing economic benefits. This study designed an apple grading system for in-field sorting, consisting of a low-cost wide-field camera, bar-shaped light sources, screw conveyors, and a computer system. An automatic method for estimating the size of moving apples based on two primary steps was developed: (1) detecting the maximum cross-sectional diameter of the apple based on its orientation, and (2) using a 3D projection correction algorithm to adjust the detected diameter. First, each apple was tracked, and its stem/calyx region was detected to determine its orientation. Then, the maximum cross-sectional diameter of the apple in the 2D image was obtained based on orientation. Finally, 3D correction equation was constructed to calibrate the diameter, addressing the projection distortion caused by the apple being away from the camera. Sixty ‘Fuji’ and sixty ‘Golden Delicious’ apples were used for the experiments. The orientation detection accuracy of the algorithm was 93.1% for ‘Fuji’ apples and 98.5% for ‘Golden Delicious’ apples. For size estimation, the percentage of apples with an error within ± 5 mm was 98.3% (‘Fuji’) and 100% (‘Golden Delicious’). Furthermore, the percentage of apples with an error within ± 2 mm was 75% (‘Fuji’) and 91.7% (‘Golden Delicious’), with an overall RMSE of 1.62 mm. The experimental results indicate that the proposed method can accurately estimate the size of apples despite changes in their posture and position during movement. This performance surpasses that of similar studies and provides technical support for apple size grading. The method also shows potential for application in sorting other near-spherical fruits and vegetables.