NIRS identification of cashmere and wool fibers based on spare representation and improved AdaBoost algorithm
Near-infrared (NIR) spectroscopy is essential for distinguishing cashmere from wool. It is fast and non-destructive. Both cashmere and wool contain keratin. Their NIR spectral images are very similar. This makes it hard to tell them apart. This article proposes a method to identify and classify two...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
De Gruyter
2025-07-01
|
| Series: | AUTEX Research Journal |
| Subjects: | |
| Online Access: | https://doi.org/10.1515/aut-2025-0047 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Near-infrared (NIR) spectroscopy is essential for distinguishing cashmere from wool. It is fast and non-destructive. Both cashmere and wool contain keratin. Their NIR spectral images are very similar. This makes it hard to tell them apart. This article proposes a method to identify and classify two similar fibers. It uses NIR spectroscopy combined with chemometrics. The chemometrics model mainly uses sparse representation and an improved AdaBoost classifier. The sparse representation is used for feature extraction. It expands the distance between sample spectra. This strategy first standardizes the dataset. It then uses the K-singular value decomposition algorithm to learn a low-dimensional dictionary. Finally, it maps the spectra using the dictionary. This approach creates a low-dimensional positive–negative distribution of sample features. It aims to widen the gap for later classification. Also, due to the many cashmere and wool species, some intra-species gaps are larger than the inter-species ones. This has increased misclassification errors. This article uses AdaBoost to assign weights to samples of different species. It optimizes these weights with many decision tree (DT) classifiers. It then uses the sparrow optimization algorithm. It finds the best number and depth of DTs. In the comparison experiments of this article, the sparse representation can effectively amplify the gap between cashmere and wool than principal component analysis and independent component analysis, and the classification efficiency of AdaBoost is also higher than the classification efficiency of K-nearest neighbors, Random Forest, and other classifiers. The combination of these two algorithms can achieve a classification accuracy of 97.4%, which can effectively classify cashmere and wool fibers. |
|---|---|
| ISSN: | 2300-0929 |