A real-time detection method and system for automobile windshield coating
By optimizing the detection model and network architecture, the problems of weak anti-interference ability, low detection accuracy and poor adaptability in automotive windshield adhesive detection have been solved, achieving high-precision, real-time adhesive detection and supporting rapid adaptation and low-cost deployment for multiple vehicle models.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA FAW CO LTD
- Filing Date
- 2026-05-28
- Publication Date
- 2026-07-17
AI Technical Summary
Existing automotive windshield adhesive detection technology is susceptible to changes in light and interference from sponge strips, making it difficult to distinguish between the first and last adhesive strips and the target adhesive strip. This results in a high false detection rate and an inability to balance real-time performance and accuracy, failing to meet the stringent requirements of international standards.
We employ a backbone and neck network combined with a channel attention mechanism, train the detection model through a two-stage cached data augmentation strategy, integrate dynamic soft label allocation strategy and combined loss function to optimize feature extraction and model training, and combine large kernel depthwise separable convolution and channel attention mechanism to improve detection accuracy and adaptability.
It enables accurate detection of adhesive line parameters and defects in complex environments, reduces false detection rate, improves detection accuracy to 99.6%, meets the real-time detection needs of production lines, supports rapid adaptation to multiple vehicle models, and reduces deployment costs.
Smart Images
Figure CN122415588A_ABST