.png){kind=icon}
.png){kind=icon}
Return
Machine learning-based detection technology for electron beam and laser processing
.png){kind=icon}
PRE
AI
DOI：10.1016/j.optlastec.2026.115292.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Electron beam and laser beam processing technologies have been increasingly adopted in industrial applications, especially in the field of microfabrication. In-depth research into their processing mechanisms, diagnostics, and simulation has become pivotal for further technological advancement. During electron beam and laser beam processing, factors such as processing parameters and material properties are prone to causing defects in the processed images, including blurred boundaries, weakened edges, and partial loss of structural features. To ad dress this issue, this paper proposes an image segmentation model for electron beam and laser beam processing, termed Squeeze-and-Excitation Dynamic Upsampling Mask R-CNN (SE-DyMask). This model incorporates a chan nel attention mechanism, employs Synthetic Fusion Pyramid Network (S-FPN) with dynamic upsampling, and optimizes feature fusion to specifically enhance segmentation accuracy in regions with weak edges and missing features. Experimental results demonstrate that compared with the original model Mask R-CNN, the enhanced SE-DyMask algorithm achieves a 2.67% improvement in AP and a 1.06% increase in AR. Moreover, under the same experimental conditions, it demonstrates competitive performance compared to established architectures such as U-Net and SegFormer, achieving 1% to 3% higher IoU on challenging microfabrication images charac terized by low contrast and fragmented structures. This research provides effective support for both academic research and industrial technological upgrading in the field of electron beam and laser beam processing.
Keywords:
Image segmentation
Electron and laser beam processing
SE-DyMask
Microfabrication
Feature fusion optimisation
Journal
O
IF:
5
Papers:
1.9K
Citations:
3.5W
