AI framework maps seagrass habitats using sonar imagery with weak supervision

Researchers have developed a weakly supervised semantic segmentation framework to map seagrass habitats using side-scan sonar imagery. This method learns pixel-level maps from image-level labels alone, employing a Vision Transformer (ViT)-based encoder-decoder and a classification branch. The framework refines pseudo-labels using a dense conditional random field and an iterative self-training scheme to handle class imbalance and noise. Experiments showed that the Lovász-Softmax loss function was most effective, and the model achieved an mIoU of 87.6% without pixel-level labels, with self-supervised pretraining further improving performance. AI

IMPACT Enables more efficient and scalable ecological monitoring by automating the analysis of sonar imagery.

RANK_REASON The cluster contains an academic paper detailing a new methodology for image segmentation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• Lovász-Softmax loss
• ScienceCast
• Side-scan sonar
• ViT
• Seagrass

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →