Analysis and Application of Multispectral Data for Water Segmentation Using Machine Learning

Monitoring water is a complex task due to its dynamic nature, added pollutants, and land build-up. The availability of high-resolution data by Sentinel-2 multispectral products makes implementing remote sensing applications feasible. However, overutilizing or underutilizing bands the product can lead inferior performance. In this work, we compare performances ten out thirteen available in for segmentation using eight machine learning algorithms. We find that shortwave-infrared (B11 B12) are most superior segmenting bodies. B11 achieves an overall accuracy $$71\%$$ while B12 $$69\%$$ across all algorithms on test site. also Support Vector Machine (SVM) algorithm favorable single-band segmentation. SVM tested over given Finally, demonstrate effectiveness choosing right amount data, use only reflectance train artificial neural network, BandNet. Even with basic architecture, BandNet proportionate known architectures semantic segmentation, achieving 92.47 mIOU requires fraction time resources run inference, making it suitable be deployed web monitor bodies localized regions. Our codebase at https://github.com/IamShubhamGupto/BandNet .