Page Banner

United States Department of Agriculture

Agricultural Research Service

Title: Gabor feature-based apple quality inspection using kernel principal component analysis

item Zhu, Bin
item Lu, Jiang
item Luo, Yaguang - Sunny
item Tao, Yang

Submitted to: Journal of Food Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/9/2007
Publication Date: 7/1/2007
Citation: Zhu, B., Lu, J., Luo, Y., Tao, Y. 2007. Gabor feature-based apple quality inspection using kernel principal component analysis. Journal of Food Engineering. 81(4):741-749.

Interpretive Summary: The apple industry has a high demand for automated apple defect sorting, because this technology significantly improves process efficiency and reduces labor cost. However, the difficulty in accurately differentiating naturally occurring apple stem and calyx ends from true defects using traditional two-dimensional, near-infrared imaging presents a major technical challenge for the apple industry. In this research, we introduced a new digital imaging processing technology by using a principal component analysis method to differentiate apple defects from stem/calyx ends. This method significantly improved the accuracy in sorting apples with true defects. The implementation of this new process will improve machine vision technology so that it can be used successfully, enabling apple growers and shippers to accurately and efficiently sort out defective apples and provide high quality products to consumers.

Technical Abstract: Automated inspection of apple quality involves computer recognition of good apples and blemished apples based on geometric or statistical features derived from apple images. This paper introduces a Gabor feature-based kernel, principal component analysis (PCA) method; by combining Gabor wavelet representation of apple images and the kernel PCA method for apple quality inspection using near-infrared (NIR) imaging. First, Gabor wavelet decomposition of whole apple, NIR images was employed to extract appropriate Gabor features. Then, the kernel PCA method, with polynomial kernels, was applied in the Gabor feature space to handle nonlinear separable features. The results show the effectiveness of the Gabor-based kernel PCA method in terms of its absolute and comparative performances in comparison to the PCA, kernel PCA with polynomial kernels, Gabor-based PCA and the SVM methods. Using the proposed Gabor kernel PCA eliminated the need for local feature segmentation, and also resolved the nonlinear separable problem. An overall 90.6% recognition rate was achieved.

Last Modified: 08/22/2017
Footer Content Back to Top of Page