.png){kind=icon}
.png){kind=icon}
Return
StarCrack: A lightweight crack segmentation network via star operation and wavelet energy
.png){kind=icon}
PRE
AI
R
G
Y
B
X
DOI：10.1016/j.measurement.2026.122680.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• A parameter-free wavelet energy module enhances shallow crack features. • Star-Wavelet Blocks at 1/4 and 1/8 stages preserve fine crack cues. • CrackSegHead uses half-resolution stems and strip convolutions for cracks. • BCE, Log-Cosh Dice, and Focal losses improve imbalanced crack learning. • StarCrack achieves 82.39% average mIoU with only 2.765 M parameters.
Keywords:
Crack segmentation
Lightweight network
Wavelet directional energy
Direction-aware feature enhancement
Pixel-unit morphology analysis
Journal
IF:
5.6
Papers:
1.9W
Citations:
5.4W
Organization
Cited Papers