Location: Appalachian Fruit Research Laboratory
Title: Identifying apple surface defects using principal components analysis and artifical neural networks Authors
|Bennedsen, Bent -|
|Peterson, Donald -|
Submitted to: American Society of Agricultural and Biological Engineers
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: October 1, 2007
Publication Date: December 1, 2007
Citation: Bennedsen, B.S., Peterson, D.L., Tabb, A. 2007. Identifying apple surface defects using principal components analysis and artifical neural networks. American Society of Agricultural and Biological Engineers. 50(6):2257-2265. Interpretive Summary: A study was done to determine if surface defects on fruit (such as bruises and insect damage) could be detected from digital images. The objective was to develop methods to detect these defects for application on automated fruit sorters in packing facilities. The method for processing the digital images developed during the work detailed in this paper included first, dimensionality reduction via Principal Components Analysis (PCA) and then, classification into defect or non-defect classes by neural networks. The digital images used were in the near-infrared range, which allowed certain types of defects to be seen more easily than in visible light images. While the success rate of 79 percent defects detected is not suitable for commercial implementation on a sorting line in a packing facility, we concluded the method developed in this paper may be useful as part of a combination of methods for defect detection.
Technical Abstract: Artificial neural networks and principal components were used to detect surface defects on apples in near-infrared images. Neural networks were trained and tested on sets of principal components derived from columns of pixels from images of apples acquired at two wavelengths (740 nm and 950 nm). In an iterative process, different ways of preprocessing images prior to training the networks were attempted. Best results were obtained by removing the background and applying a Wiener filter to the images. Overall, the best performance obtained was 79 percent of the defects detected in a test set consisting of 185 defect.