Problem
Research questions and friction points this paper is trying to address.
Detects boundaries in poly-crystalline microscopic images
Eliminates defects in raw microscopic images
Preserves geometric and topological characteristics of grains
WPU-Net: Boundary learning by using weighted propagation in convolution network
🤖 AI Summary
To address the challenges of blurred and inaccurately localized grain boundaries in polycrystalline microstructural images, this paper proposes a boundary-aware weighted propagation mechanism. The core innovation is a learnable Weighted Propagation Unit (WPU) that explicitly embeds boundary priors into the convolutional architecture, enabling end-to-end modeling of spatial propagation weights for boundary pixels. The method further integrates multi-scale features with orientation-aware weight maps to enhance boundary discrimination. Evaluated on standard benchmarks including BSDS500, the approach achieves significant improvements in boundary detection accuracy—yielding an F-score gain of over 3.2% compared to state-of-the-art methods such as HED and RCF. This advancement provides a more robust and precise solution for grain boundary segmentation, thereby facilitating quantitative analysis of material microstructures.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Uses U-Net based Weighted Propagation Convolution Network
Introduces spatial consistency to eliminate image defects
Customizes adaptive boundary weights for each grain pixel
🔎 Similar Papers
No similar papers found.
