Comparison of Principal Component Analysis and different band selection methods for classification of construction waste with hyperspectral images

Lennard Wunsch, Gunther Notni

Abstract

This work presents a machine learning pipeline for construction waste sorting utilizing spectral imaging and comparing different dimensionality reduction methods. Aiming to correctly classify over 90% of objects in our dataset, we applied band selection methods based on Mutual Information, Fisher’s Score, Sequential Forward Selection, and Sequential Backward Selection. In addition, we examined Principal Component Analysis (PCA) and Categorical Maximum Spectral Difference. The performance of each pipeline is evaluated using metrics such as accuracy, precision, recall, F1 Score, and AUC-ROC.

Download

IMEKO-TC2-2025-001.pdf

DOI

10.21014/tc2-2025.001

IMEKO TC

TC2 - Photonics

Back to the proceedings