Skip to main content
ARS Home » Northeast Area » Kearneysville, West Virginia » Appalachian Fruit Research Laboratory » Innovative Fruit Production, Improvement, and Protection » Research » Publications at this Location » Publication #164737
Title: PERFORMANCE OF A SYSTEM FOR APPLE SURFACE DEFECTS IDENTIFICATION IN NEAR INFRARED IMAGES

Author
item BENNEDSEN, BENT - THE ROYAL VET & AG UNIV
item Peterson, Donald

Submitted to: Biosystems Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/3/2004
Publication Date: 2/12/2005
Citation: Bennedsen, B.S., Peterson, D.L. 2005. Performance of a system for apple surface defects identification in near infrared images. Biosystems Engineering. (2005) 90(4). 419-431 doi: 10.1016/j.biosystemseng.2004.12.005 PH-Postharvest Technology. 1537-5110

Interpretive Summary: In order to ensure that consumers have access to high and consistent quality apples, automatic on line sorting according to surface defects is required. In this work, the combined performances of four different methods for identifying defects in images of apples were quantitatively tested, using eight apple varieties. The detection system successfully identified an average of 87.7% of all the defects presented to the system. Detection of bruises ranged from 85 to 100%; other defects from 66 to 80%. The performance of the system is such that it could potentially be implemented as it is for inspection of apples for processing. For fresh market apples, slight improvements in the performance will be required.

Technical Abstract: This paper reports the development and testing of machine vision systems for sorting apples for surface defects, including bruises. The system operated on apples, which were oriented with the stem/calyx axis perpendicular to the imaging camera. Grey scale images in the visible wavebands were used to verify orientation. Images for detection of defects were acquired through two optical filters at 740 nm and 950 nm, respectively. Defects were detected using a combination of three different threshold segmentation routines and one routine based on artificial neural networks and principal components. The paper reports quantitative measurement of the performance of the system for verification of orientation and a combination of the four segmentation routines. The routines were evaluated using eight different apple varieties. The ability of the routines to find individual defects and measure the area ranged from 77 to 91% for the number of defects detected, and from 78 to 92.7% of the total, defective area.