分割
光学
材料科学
钻石
图像分割
镜头(地质)
人工智能
复合数
计算机视觉
算法
计算机科学
图像处理
图像(数学)
图像质量
作者
Yuheng Wang,Yabing Wang,Yuhai Gu,Weiwei Zhang
出处
期刊:
[SPIE]
日期:2026-04-09
卷期号:: 34-34
摘要
The fabrication of diamond compound refractive lenses (CRLs) relies on machine vision and femtosecond laser technology for center localization and precision machining, in which accurate image segmentation plays a critical role in error compensation. To overcome the limitations of conventional models in segmentation accuracy and computational efficiency, we propose an improved U-Net model that integrates MobileNetV2 as the encoder for lightweight and efficient feature extraction, and incorporates Efficient Channel Attention mechanism to enhance feature representation. Experiments on both a self-constructed dataset and an independent test set demonstrate that the improved U-Net model achieves higher segmentation accuracy and efficiency with significantly reduced memory consumption compared to conventional models. Furthermore, the improved U-Net model has been successfully applied to the practical fabrication of diamond CRLs at the Beijing Huairou Synchrotron Radiation Facility, confirming its robustness and practical value.
科研通智能强力驱动
Strongly Powered by AbleSci AI