基于卷积神经网络的混凝土裂缝无损检测量化分析方法
By generating a semantically mutually exclusive mask for cracks and a virtual orthophoto projection, and eliminating non-coplanar feature points, the problem of insufficient accuracy in monocular vision crack detection is solved, and high-precision crack width quantization and a simplified detection process are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGXI ZHUANG AUTONOMOUS REGION CONSTR ENG QUALITY INSPECTION CENT CO LTD
- Filing Date
- 2026-01-12
- Publication Date
- 2026-07-17
AI Technical Summary
Existing monocular vision crack detection technology suffers from insufficient accuracy in quantifying crack physical width under non-perpendicular shooting conditions due to interference from non-planar features inside the crack and perspective distortion effects. Furthermore, traditional methods rely on external sensors or calibration references, which increases the complexity and difficulty of the detection system.
By using a convolutional neural network-based method, a semantically mutually exclusive mask for cracks is generated, non-coplanar feature points are eliminated, and a virtual orthophoto projection is constructed using geometric verification and plane normal vectors. Combined with Zernike moments for sub-pixel localization, high-precision quantization without external devices is achieved.
It improves the accuracy and consistency of crack width quantification, simplifies the detection process, reduces the requirements for the posture of the imaging equipment, and achieves sub-pixel level resolution for detecting tiny cracks.
Smart Images
Figure CN121837246B_ABST