四川大学学报（自然科学版）2025，Vol.62Issue(2)：388-398,11.DOI:10.19907/j.0490-6756.240082
基于深度学习的硬质合金图像分割算法
Deep learning-based image segmentation algorithm for tungsten carbide
摘要
Abstract
A deep learning network segmentation algorithm based on a multi-level attention fusion mecha-nism is proposed to extract the tungsten carbide grain regions in micrographs of hard alloy.Utilizing the UNet network model,a multi-level attention fusion module is incorporated within the skip connections between deep down-sampling and up-sampling layers to enhance the feature representation of tungsten carbide grains in down-sampled feature maps.By learning the weights for each channel,this mechanism accentuates the grain features,thereby enhancing the model's focus on crucial feature channels.This strengthens the correla-tion between features and improves the model's capability to represent input data.Comparative and ablation experiments are conducted on various types of hard alloy test sets,demonstrating the effectiveness of the pro-posed approach in segmenting micrographs of hard alloys.关键词
硬质合金显微图像/图像分割/对比度/深度学习/自注意力机制Key words
Hard alloy micrograph/Image segmentation/Contrast/Deep learning/Self-attention mechanism分类
信息技术与安全科学引用本文复制引用
阮鹏,何小海,滕奇志..基于深度学习的硬质合金图像分割算法[J].四川大学学报（自然科学版）,2025,62(2):388-398,11.基金项目
国家自然科学基金(62071315) （62071315）
