Borrowing least squares analysis from spectral unmixing to classify plastics in SWIR hyperspectral images

Plastics have long been receiving attention due to their abundance in daily use, as well as their loss to the environment as debris. Plastic pollution is widely accepted as an environmental crisis, particularly in marine environments as millions of tons of plastics enter the oceans annually. Although some macro plastics can be determined using visible-range or VNIR hyperspectral imaging, microplastics as well as those that are colorless or have similar pigmentation are difficult to differentiate in the visible spectral regions. SWIR or short-wave infrared hyperspectral imaging offers a solution for plastics detection in the near infrared spectrum. This study builds on a recent work for detection and identification of plastics using classical feature extraction techniques and spectral indices. Here, we apply least squares analysis borrowed from linear spectral unmixing methods for the classification of plastics from SWIR hyperspectral data. In this research, we compare the results of the two approaches. The two methods produce similar results even though the first approach only utilizes a limited number of features, and the second approach makes use of the entire spectral bands represented in each scene pixel.