Using a machine learning-based hyperspectral image classification method for stray light pollution level assessment

Hyperspectral images (HSIs) can suffer essential information loss when the hyperspectral imaging system is affected by stray light interference. This diminishes the advantages of HSIs, which could otherwise distinguish different objects by utilizing both spatial and spectral data. To facilitate further applications, this paper proposes a classification-based model that can evaluate the impact of stray light pollution on HSIs. In this model, an HSI is first classified by a weighted spatial-aware cooperative classifier after super pixel segmentation. By comparing the classification results of a contaminated hyperspectral image (HSI) and its stray light-free counterpart on a pixel-wise basis, the pollution degree can be quantitatively evaluated based on changes in classification confidence. A pollution level distribution map is finally generated, which intuitively illustrates the impact of stray light pollution on hyperspectral data. This assessment scheme offered insights for evaluating the degree of stray light pollution and can be extended to other hyperspectral datasets for different application tasks.