|Cheng, Xuemei - U.MD.,COLLEGE PARK|
|Tao, Yang - U.MD.,COLLEGE PARK|
Submitted to: Transactions of the ASAE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: May 6, 2004
Publication Date: July 1, 2004
Citation: Cheng, X., Chen, Y.R., Tao, Y., Wang, C.Y., Kim, M.S., Lefcourt, A.M. 2004. A novel integrated PCA and FLD method on hyperspectral imaging feature extraction method for cucumber chilling damage inspection. Transactions of ASAE. 47(04):1313-1320. Interpretive Summary: Much research has been concentrated on feature extraction and dimensionality reduction of hyperspectral imaging data set to a single discriminate function for classification purposes. Principal component analysis (PCA) and linear discriminant analysis (LDA) are two well-known linear transforms or projection methods for feature extraction and dimensionality reduction. PCA is commonly used in fecal-contaminated apple inspection and other fruit and vegetable quality and safety inspection applications, but sometimes fails to yield good classification since PCA is better for representation than classification. PCA is not necessarily good at drawing distinctions between patterns. On the other hand, Fisher's Linear Discriminant (FLD) method is often used in developing the criteria for selecting spectral bands for classification, but is susceptible to outliers and noise. The proposed new method combines the PCA and FLD methods for feature extraction and dimensionality reduction. It was applied to the classification of cucumbers by damage due to chilling injury. This new integrated PCA-FLD method was shown to yield better classification performance than the individual PCA and FLD methods, when tested on various types of samples. This method is ready to be extended to other hyperspectral imaging applications for fruit and vegetable safety and quality inspections. Researchers can apply this technique for obtaining optimal discriminate function for classifying produce for safety and quality.
Technical Abstract: High-resolution hyperspectral imaging (HSI) provides an abundance of spectral data for feature analysis in image processing. Usually, the amount of information contained in hyperspectral images is excessive and redundant, and data mining waveband selection is needed. In applications for fruit and vegetable damage inspection, effective spectral combination and data fusing methods are required in order to select a few optimal wavelengths without losing key information in the original HSI data. In this paper, we present a new method that combines principal component analysis (PCA) and Fisher's Linear Discriminant (FLD) method in a way that maximizes the representation and classification effects on the extracted new feature bands. The method is applied to the detection of chilling injury on cucumbers. Compared with PCA and FLD when used separately, this new integrated PCA-FLD method has achieved results showing better classification performance when tested on different types of samples. This method is ready to be extended to other hyperspectral imaging applications for fruit and vegetable safety and quality inspections.