Wasimul Karim, Nur Mohammad Fahad, Abdul Hasib Siddique +4 more
Recycling an electrolyzer means telling its materials apart well enough for a machine to take it to pieces. That is harder than it sounds, because the materials look alike, overlap spectrally, come in irregular shapes, and appear in wildly unequal amounts.
Severe class imbalance is the quiet difficulty. When one material dominates the image, a model can score well overall while missing the scarce component entirely, and the scarce component is frequently the valuable one that makes recycling worth doing.
HREM-Net combines hyperspectral data with ordinary RGB in a dual-branch design, which is a sensible answer to the specific problem: two materials that look identical to a camera often differ in how they reflect wavelengths the eye cannot see. Where colour cannot separate them, spectrum can.
Accurate segmentation of electrolyzer materials is essential for automated disassembly, sustainable recycling, and circular manufacturing in hydrogen technologies. However, this task is challenging due to strong visual similarity between materials, spectral overlap, irregular shapes, and severe class imbalance. To address these challenges, we propose an AI-driven dual-branch framework, Hyperspectral-RGB Electrolyzer Materials Network (HREM-Net), that combines hyperspectral…
ToolSciVer: Multimodal Scientific Claim Verification with Visual Tool Augmented Reinforcement Learning
arXiv (cs.AI) · July 17, 2026A Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governance
arXiv (cs.CV) · July 17, 2026Toward Semantic Communication for Real-time Mobile 3D Reconstruction
arXiv (cs.AI) · July 17, 2026CRAFT: Clustering Rubrics to Diagnose Weak LLM Capabilities and Generate Targeted Fine-Tuning Data