A segmentation method for segmenting high-density microreactor images
By combining frequency domain preprocessing and U-Net network with geometric vector field branching, the problems of size perturbation and adhesion in high-density microreactor image segmentation are solved, achieving high-precision and robust image segmentation results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI INSTITUTE OF INFECTIOUS DISEASE & BIOSECURITY
- Filing Date
- 2026-04-10
- Publication Date
- 2026-07-17
AI Technical Summary
Existing image processing algorithms struggle to achieve accurate segmentation in high-density microreactor scenarios, especially under complex fluorescence backgrounds and noise interference. Traditional methods suffer from quantitative errors and lack adaptability to different imaging environments.
Frequency domain preprocessing combined with U-Net network is used to suppress background noise and enhance the contour features of microreactors through high-pass filter. Geometric vector field branch is introduced for foreground localization and geometric topology-guided segmentation. Cosine similarity loss and average diameter constraint are used to achieve accurate segmentation.
It effectively resists size disturbances and adhesion crosstalk, enabling precise segmentation of high-density microreactor images, improving segmentation accuracy and robustness, and adapting to microreactors of different shapes and sizes.
Smart Images
Figure CN122415633A_ABST