This is the authors' abstract. We don't add key points for this paper.
This study presents an innovative segmentation algorithm that seamlessly integrates an adaptive multi-scale attention mechanism with the Lossab loss function to tackle challenges like overlapping molten pool contours and arc interference during image segmentation. Initially, an adaptive multi-scale channel attention mechanism is introduced and fused with a feature pyramid, forming a multi-scale attention pyramid structure that significantly enhances the network’s capability to extract intricate molten pool features. Subsequently, a tailored Lossab loss function is devised to boost the network’s precision in detecting the edges of the molten pool. Experimental outcomes reveal that the proposed method attains a remarkable segmentation accuracy of 96.77% within the molten pool region on the TIG molten pool image dataset. The AMSNet algorithm outperforms PIDNet by 2.51% and SegNet by 4.66% in terms of molten pool segmentation accuracy, providing solid evidence of its effectiveness in tackling the task of molten pool image segmentation. The model performs well in computational efficiency, data processing and complexity balance, making it suitable for industrial field deployment.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: