Application of principle component analysis (PCA) in pattern recognition

Principle component analysis (PCA) is an useful multivariate technique widely applied in similarity comparison among study units. To date, there are many publications available related to insect numerical classification and insect ecological studies. Unfortunately very few publications are focused on the application of PCA in pattern recognition and model prediction. The first objective is developing an algebra procedure to update group statistics in discriminant analysis after absorbing new study units. The second objective is discussing confidence eclipse region computation to delineate the confidence region boundary of this specific group by Hotelling’s T2 test criterion by assumptions that the principle components are orthogonal and normal distributed, which give visual standards to decide acceptance or rejection of an investigation unit. If more than two principle components should be considered in discriminant analysis, a set of numerical criteria is developed instead of confidence eclipse boundary. The third objective is discussing the criterion to judge how many principle components should be kept to correctively show the numerical relationships among study units, and simultaneously keep the balance between the simplicity and precision. Finally, with survey data of citrus leaf and fruit, and soil nutrition elements concentrations collected from more than 30 orange (Citrus sinensis) gardens in South Jiangxi, the application of this PCA computation technique in similarity comparison and pattern recognition is also discussed in this paper.