Wool/cashmere identification based on projection curves

Abstract

Using compound microscopy is one of the major options for the identification of cashmere/wool. To interpret human perception via machine vision, microscopic images captured by a charge-coupled device camera were transferred into projection curves. Three different deciphering methods, recurrence quantification analysis, direct geometrical description, and discrete wavelet transform were employed to reveal the embedded numerical features. The extracted parameters were used to screen the supervised classification methods, including a neural network with multilayer perceptrons, kernel ridge regression/classification, and the support vector machine (SVM). The experimental results indicated that the proposed projection curves could be used as a mathematical replica in automatic cashmere/wool identification. The best accuracy came from a SVM-trained decision function with the parameters extracted from recurrence quantification analysis.