Submitted to: Transactions of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/2/2013
Publication Date: 7/9/2013
Citation: Mizushima, A., Lu, R. 2013. A low-cost color vision system for automatic estimation of apple fruit orientation and maximum equatorial diameter. Transactions of the ASABE. 56(3):813-827.
Interpretive Summary: In-field presorting is intended to separate fresh market-grade apples from inferior apples that are only suitable for processed products (i.e., juice, applesauce, canning, dried, etc.), so that growers can achieve cost savings in postharvest storage and packing, and packinghouses can have better inventory management. Hence the overall goal of this research was to develop technology for automatic in-field presorting and segregation of undersized and defective fruit. To achieve this goal, a low-cost machine vision system was built for in-field presorting of apples, which consisted of a color digital camera, energy-saving LED lights and a two-lane conveyor with bi-cone rollers. Image processing algorithms were developed for estimation of apple shape, orientation and size. Laboratory tests of the machine vision system were conducted for 'Delicious' (D), 'Empire' (EM), 'Golden Delicious'(GD), and 'Jonagold' (JG) apples at a speed of four fruit per second. The orientation estimation algorithm showed more than 85% success rate for elongated D and GD apples, whereas it had less than 70% success rate for round-shaped EM and JG apples. The machine vision system achieved superior fruit size estimations with the mean error of 1.79 mm, and 98% or more of the test apples were estimated within ±5 mm. The system had 4.3% classification error for undersized apples (i.e., less than 63.5 mm in size), versus 15.1% error for a commercial mechanical sizing machine. The low-cost hardware configuration would meet the required presorting speed of 4-6 apples per second and allow implementing additional color grading and defects detection functions for the in-field presorting system.
Technical Abstract: The overall objective of this research was to develop an in-field presorting and grading system to separate undersized and defective fruit from fresh market-grade apples. To achieve this goal, a cost-effective machine vision inspection prototype was built, which consisted of a low-cost color camera, LED (light-emitting diode) lights and a generic bi-cone conveyor. Algorithms were developed for image distortion corrections and for real-time estimation of apple orientation, shape and size. The machine vision system was tested and evaluated for 'Delicious'(D), 'Empire'(EM), 'Golden Delicious'(GD), and 'Jonagold'(JG) apples at a speed of four fruit per second. The orientation estimation algorithm had 87.6% and 86.2% accuracies for D and GD apples, respectively, within ±20°of actual fruit orientation, whereas it performed less satisfactorily for round-shaped EM and JG apples. The machine vision system achieved good fruit size estimations with the overall root mean square error of 1.79 mm for the four varieties of apple, and it had a two-size grading error of 4.3%, versus 15.1% by a mechanical sizing machine. The system provides a cost effective means for sorting apples for size.