Dataset and software for processing of hyperspectral images of different CDW materials

Overview

The provided scripts are designed to process hyperspectral images of construction and demolition waste (CDW) materials, extract relevant features, and train a machine-learning model for material classification. The scripts perform the following tasks:

1. Feature Extraction: Extract spectral features from hyperspectral data.
2. Background Removal and Subset Extraction: Remove backgrounds from images and extract subsets for analysis.
3. Data Visualization: Generate plots to visualize the extracted features and reflectance curves.
4. Machine Learning Model Training: Using the extracted features, train and evaluate a multilayer perceptron (MLP) classifier.

Prerequisites

Before running the scripts, ensure that you have the following:

• Python 3.x installed on your system.
• Required Python packages:
  • numpy
  • matplotlib
  • scipy
  • pandas
  • scikit-learn
  • seaborn
  • rembg (for background removal)
  • Pillow (PIL)
• Hyperspectral data files in .mat format containing calibrated hyperspectral cubes and wavelength information.
• A directory structure to organize input and output files as described in each script.

Scripts Description

1. hyperspectral_features_v2.py

Purpose

This script processes individual hyperspectral image files to extract spectral features from a central subset of the image. It generates RGB images from the hyperspectral data, plots the mean reflectance spectra, and outputs a LaTeX-formatted table containing the extracted features.

Functionality

• Loading Data: Reads .mat files containing hyperspectral data from a specified input directory.
• Feature Calculation:
  • Calculates mean reflectance within a central window of the image.
  • Extracts spectral features such as peak wavelength and area under the reflectance curve.
  • Records reflectance values at selected wavelengths, including standard RGB channels and additional wavelengths.
• RGB Image Generation: Creates RGB images using specific wavelengths corresponding to the red, green, and blue channels.
• Spectra Plotting: Plots the mean reflectance spectra for each sample.
• LaTeX Table Generation: Produces a LaTeX-formatted table of the extracted features for inclusion in a report or paper.

Usage Instructions

1. Prepare Input Data:

   • Place your .mat files containing the hyperspectral data in the appropriate input directory (e.g., input/mortar).

2. Run the Script:

   • Modify the materials list at the end of the script to include the materials you want to process (e.g., materials = ['mortar']).
   • Execute the script:
     bash