New Classification of Textile Samples through lp Norm Spectral Enhancement Using Template Filters Combining the Analytic Geometry Technique

Abstract

This paper introduces a novel method for classifying textile fibers into three groups: natural fibers, synthetic fibers, and blended fibers using near-infrared (NIR) spectra obtained via the NeoSpectra-Micro sensor. Our approach involves preprocessing and employing the l_p-norm with p = ∞, 1, and 2 to enhance spectral signals. These enhanced signals alongside textile template filters were obtained from both natural and synthetic fiber groups. Next, the template filters are used to construct a new 2x1 feature vector through covariance-based techniques to effectively reduce spectral data dimension. The feature vector is pivotal for establishing two threshold lines together with an analytical geometry technique to classify for accurate fiber groups. To evaluate the performance of the proposed method, experiments were conducted by using three groups of fiber samples: 210 natural fiber spectra, 480 synthetic fiber spectra, and 270 blended fiber spectra. The dataset was divided into training and testing sets with ten random iterations exploring eight ratios and l_p-norm enhancements for training and evaluation. Remarkably, the experimental result has shown that the overall accuracy remains consistent across the three cases of the l_p-norm enhancements providing the similar accuracies. Considering the limited computational resource, the l₁-norm emerges as a practical choice for embedded systems, emphasizing its practicality for implementation. Moreover, the proposed method additionally provides high accuracies (mean ± standard deviation) of 0.9995 ± 0.0006, 0.9999 ± 0.0004, and 0.9999 ± 0.0005, whereas the ratio of the train and test data is equal to three cases: 10:90, 20:80 and 30:70, respectively, and achieves an exceptional overall accuracy of 100%, whereas the ratio of the train and test data is equal to five cases: 40:60, 50:50, 60:40, 70:30 and 80:20.