Filter-Interfered FTIR Spectra Dataset for Microplastic Classification

Purpose and Context The escalating challenge of microplastic pollution necessitates robust and reliable identification methods. Fourier-transform infrared (FTIR) spectroscopy is a widely adopted technique for microplastic analysis. However, a crucial step in sample preparation for aquatic environments involves depositing microplastics onto membrane filters, which often introduce spectral interference during FTIR acquisition. Many common membrane filters exhibit strong infrared absorption that overlaps with key characteristic bands of microplastic polymers, distorting spectra and complicating accurate classification. This dataset and accompanying code address this challenge by providing resources for developing and evaluating microplastic classification models in the presence of such filter-induced spectral interference. Dataset Description This dataset, named "Dataset of Filter-Interfered FTIR Spectra.csv," comprises 1320 FTIR spectra designed for microplastic classification in scenarios with membrane filter interference. Content: The dataset includes spectra from 22 distinct microplastic types, with 60 spectra per type. The microplastic types are: Acrylic, Cellulose, Epoxidized Natural Rubber (ENR), Ethylene Propylene Diene Monomer (EPDM) Rubber, High-Density Polyethylene (HDPE), Low-Density Polyethylene (LDPE), Nylon, Polybutylene Adipate Terephthalate (PBAT), Polybutylene Succinate (PBS), Polycarbonate (PC), Polyether Ether Ketone (PEEK), Polyetherimide (PEI), Polyethylene Terephthalate (PET), Polylactic Acid (PLA), Poly(methyl methacrylate) (PMMA), Polyoxymethylene (POM), Polypropylene (PP), Polystyrene (PS), Polytetrafluoroethylene (PTFE), Polyurethane (PU), Polyvinyl Alcohol (PVA), and Polyvinyl Chloride (PVC). Acquisition: Spectra were acquired using a Nicolet iS5 spectrometer (iD7 base, Thermo Scientific, USA) operating in Attenuated Total Reflectance (ATR) mode. Each spectrum was recorded over a wavenumber range of 650-3999.5 cm$^{-1}$ at a resolution of 0.5 cm$^{-1}$. Interference Source: The acquired FTIR spectra are intentionally distorted by spectral interference from a Cytiva Whatman membrane filter (pore size: 0.45 µm; diameter: 47 mm), which was present during acquisition. It is crucial to note that these spectra are not pristine, high-quality data; their characteristic microplastic bands are significantly obscured by the membrane filter's spectral signature. File Structure: The CSV file contains 1320 rows (each representing a single spectrum) and 6700 columns (each corresponding to a wavenumber). Wavenumbers start at 650 cm$^{-1}$ and increase in increments of 0.5 cm$^{-1}$ up to 3999.5 cm$^{-1}$. This is also shown in the file "Wavenumber.csv". Data Organization: Spectra are organized by microplastic type. Every 60 consecutive rows correspond to a single microplastic type, following the exact order listed above. For example, rows 1-60 are Acrylic, rows 61-120 are Cellulose, rows 121-180 are ENR, and so forth. Python Code Description This repository includes Python code in .ipynb (Jupyter/Colab notebook) format, supporting the methodology in our research for microplastic classification using dimensionality reduction and deep learning. The notebooks facilitate synthetic dataset creation, preprocessing, and classification performance evaluation. File "1. Create Synthetic Datasets.ipynb": This notebook generates five distinct synthetic datasets from the original "Dataset of Filter-Interfered FTIR Spectra.csv" file. File "2. Preprocessing Stage.ipynb": This notebook performs preprocessing steps, including baseline correction and min-max normalization, on the synthetic datasets. File "3. Classification Performance - N=40.ipynb": This notebook evaluates the classification performance of the proposed approach, which combines dimensionality reduction techniques with deep learning classification. Disclaimer The data and code are provided as is, without any warranty, for research and educational purposes.