语义分割模型训练方法、装置、计算机设备和存储介质
An improved method for constructing class prototype vectors and loss functions in loan monitoring solves the problem of inaccurate semantic segmentation under limited sample data in loan monitoring in remote areas, and achieves high-accuracy segmentation of loan monitoring images with limited samples.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INDUSTRIAL AND COMMERCIAL BANK OF CHINA
- Filing Date
- 2022-12-12
- Publication Date
- 2026-07-17
AI Technical Summary
In loan monitoring in remote areas, due to the difficulty of background investigation and the scarcity of valid monitoring data, existing semantic segmentation models based on prototype learning are not accurate enough in terms of segmentation results with limited sample data.
By acquiring sample support images and mask labels from the training task, class prototype vectors are determined, first and second loss functions are constructed, a semantic segmentation model is trained based on the target loss function, and semantic segmentation is performed with a small number of samples using an improved prototype fusion strategy.
With only a small amount of sample data, the semantic segmentation accuracy of loan monitoring images was improved, ensuring the accuracy of pre-loan investigation and post-loan monitoring.
Smart Images
Figure CN115861617B_ABST