Sailing Through Spectra: Unveiling the Potential of Multi-Spectral Information in Marine Debris Segmentation

Dyutit Mohanty, Aditya Kasliwal, Bharath Udupa, Pratinav Seth

Keywords: Semantic Segmentation, Remote sensing, Climate change, Ocean plastic, Multi spectral satellite imagery

TL;DR: Experimentation with multi-spectral features from remote sensing data, to enhance the ability of deep learning models in discerning ocean plastic waste from other ocean surface features, to help with plastic monitoring efforts.

Abstract: Plastic debris in ocean waters poses ecological and economic challenges. Addressing this issue begins with estimating plastic distribution in oceans for effective policy and awareness efforts. Traditional monitoring methods are costly and labour-intensive, with limited coverage. Deep learning models using multispectral remote sensing data show promise in overcoming these limitations. However, accurately distinguishing floating plastic from other sea surface features remains challenging. In our work, we use the multi-spectral Sentinel-2 MARIDA dataset to explore the impact of various spectral feature combinations on the performance of deep learning models for segmenting marine plastic in the presence of other sea surface features. This innovative approach improves accuracy and serves as an open benchmark for multi-spectral marine debris segmentation.

Submission Number: 244