基于无监督域适应的骨髓细胞分类方法、系统、设备及介质
By introducing high-frequency constraints of wavelet transform and a category-aware module into bone marrow cell classification, the problems of domain offset and category bias are solved, improving the multi-class classification accuracy of bone marrow cell images. In particular, higher classification results are achieved in image data migration between different institutions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTH CHINA UNIV OF TECH
- Filing Date
- 2024-04-19
- Publication Date
- 2026-07-17
AI Technical Summary
Existing unsupervised domain adaptation methods suffer from domain shift and class bias in bone marrow cell classification, resulting in poor classification performance. In particular, the classification decision boundary is blurred in multi-class classification, and it is impossible to effectively utilize bone marrow cell images from different institutions.
By adding a wavelet transform-based high-frequency constraint module and a discriminative category-aware module to the feature extractor, domain-invariant features are extracted and category decision-making is strengthened, thus constructing a domain-adaptive classification network model based on high-frequency constraints and category awareness.
It improves the accuracy of bone marrow cell image classification, especially in multi-class classification, significantly improving the classification effect between categories, reducing the false label error rate, and enhancing the network's attention to categories.
Smart Images
Figure CN118552951B_ABST