結合光譜平滑化技術的廢紡分類辨識系統設計與實現

Abstract

隨著全球環境保護（Environmental）、社會責任（Social）與公司治理（Governance, ESG）議題的重要性日益提升，紡織品的回收和再利用成為了紡織產業中極具挑戰性和急需解決的問題之一。實驗室專注於回收具經濟價值的紡織材料，包括聚酯纖維（Polyester）、尼龍（Nylon）、棉（Cotton）以及聚酯纖維與棉的混紡物。以往的材料鑑別主要依賴於化學分析技術，該方法耗時且可能對環境造成影響。因此，開發一種快速、環保且非破壞性的材料辨識方法具有重要意義。本研究構建了一基於近紅外光譜儀（NIR）及Linux系統架構的紡織品回收系統，用於廢紡材料的分類辨識。然而，由於測量環境的光場變化，其光譜量測結果存在差異，影響系統定量分析穩定性。因此本研究引入Cubic Spline（三次樣條）數據平滑化技術，將光譜解析度由10 nm提升至1 nm，增強光譜數據的細節。同時，結合標準參照與校正方法，使用五種反射強度參照標準（100%、99%、75%、50%、2%反射率）進行光譜校正，使用Python進行了數據分析與處理，將絕對強度轉換為相對強度，以消除環境光場變化對測量的影響。此外，吾人結合主成分分析（PCA）來提取光譜數據的主要特徵，並採用支持向量機（SVM）進行分類模型的建立。這些技術有效地提升了系統的穩定性和分類精度。同時標準參照與校正方法有效地校正了不同環境下的光譜偏差，提升了測量結果的一致性。實驗結果顯示，經處理後的光譜數據在樣品分類辨識中取得了較高的準確率透過這些步驟，吾人旨在對衣物的材質進行定性分析並同時在混紡類別中以定量分析進行精準識別。本研究成功設計並實現了一套結合光譜平滑化技術的廢紡分類辨識系統。這一過程對於材料鑑定和化學分析提供了重要的基礎，使吾人能夠更有效地處理和回收紡織品。通過Cubic Spline數據平滑化和標準參照校正，系統的穩定性和辨識準確性得到了顯著提升。未來將進一步優化光譜數據處理算法，並引入機器學習模型，同時，計劃拓展系統對其他紡織材料的適用性，為紡織品回收再利用提供更全面的技術支持。With the growing importance of Environmental, Social, and Governance (ESG) issues, textile recycling and reuse have emerged as one of the most challenging and urgent problems in the textile industry. This study focuses on recycling economically valuable textile materials, including polyester, nylon, cotton, and polyester-cotton blends. Traditionally, material identification has relied heavily on chemical analysis techniques, which are time-consuming and potentially harmful to the environment. Therefore, developing a rapid, eco-friendly, and non-destructive method for material identification is of great significance.In this research, we developed a textile recycling system based on Near-Infrared (NIR) spectroscopy and a Linux-based architecture for the classification and identification of waste textiles. However, variations in the light field of the measurement environment introduced inconsistencies in spectral measurements, affecting the stability of the system’s quantitative analysis. To address this issue, the study introduced Cubic Spline interpolation for spectral smoothing, improving the spectral resolution from 10 nm to 1 nm, enhancing the detail of spectral data. Additionally, a calibration method using five reference reflectance standards (100%, 99%, 75%, 50%, and 2% reflectance) was employed to convert absolute intensity to relative intensity, data analysis and processing were conducted using Python, mitigating the impact of environmental light field changes on measurements.Furthermore, Principal Component Analysis (PCA) was used to extract key features from the spectral data, and a Support Vector Machine (SVM) was employed to build the classification model. These techniques effectively improved system stability and classification accuracy. The calibration method also successfully corrected spectral deviations caused by varying environmental conditions, enhancing the consistency of measurement results. Experimental results showed that the processed spectral data achieved high accuracy in sample classification and identification.This research successfully designed and implemented a textile classification and identification system that incorporates spectral smoothing techniques. This process provides a solid foundation for material identification and chemical analysis, enabling more efficient handling and recycling of textiles. The use of Cubic Spline smoothing and reference calibration significantly improved the system’s stability and identification accuracy. Future work will focus on further optimizing spectral data processing algorithms and incorporating more advanced machine learning models to enhance the system’s automation and classification efficiency. Additionally, plans to expand the system's applicability to other textile materials will provide more comprehensive technical support for textile recycling and reuse.