AlgaeNet: A Deep-Learning Framework to Detect Floating Green Algae From Optical and SAR Imagery

This article developed a scalable deep-learning model, the AlgaeNet for floating Ulva prolifera ( U. prolifera ) detection in moderate resolution imaging spectroradiometer (MODIS) and synthetic aperture radar (SAR) images. We labeled 1055/4071 pairs of samples, among which 70%/30% were used training/validation. As result, model reached an accuracy 97.03%/99.83% mean intersection over union 48.57%/88.43% MODIS/SAR The was designed based on classic U-Net with two tailored modifications. First, physics information input multichannel multisource remote sensing data. Second, new loss function to resolve class-unbalanced samples (algae seawater) improve performance. In addition, this is expandable process images from optical sensors (e.g., MODIS/GOCI/Landsat) SAR Sentinel-1/GF-3/Radarsat-1 or 2), reducing potential biases due selection extraction thresholds during traditional threshold-based segmentation. satellite containing Yellow Sea draw conclusions. adding 10-m high-resolution imagery shows 63.66% increase algae 250-m MODIS image alone. we define submerged ratio number (FS ratio) parts detected by MODIS. A research vessel measurement confirms FS be good indicator representing different life phases .