Small sample image classification method based on reflective semantic prompting mechanism and multi-branch semantic alignment

By employing a reflective semantic prompting mechanism and a few-sample image classification method with multi-branch semantic alignment, the problems of category prototype bias and unreliable semantic information under small sample conditions are solved, achieving high-accuracy classification under extremely low sample conditions.

CN122176362APending Publication Date: 2026-06-09NANTONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Applications(China)
Current Assignee / Owner
NANTONG UNIV
Filing Date
2026-02-03
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing few-sample image classification methods struggle to accurately construct discriminative category prototypes under extremely low sample conditions. Directly introducing category names or textual semantic information can easily lead to semantic ambiguity. Cross-modal alignment methods are insufficient to fully model the differences between semantic and visual features, thus limiting the improvement of model classification performance.

Method used

A reflective semantic prompting mechanism is adopted to generate category semantic representations that are both stable and discriminative. A multi-branch semantic alignment module is used to achieve deep alignment and fusion of semantic information and visual features in multiple subspaces, thereby constructing a stable and discriminative category prototype.

Benefits of technology

It improves the accuracy of class prototype representation under extremely low sample conditions, enhances the model's ability to characterize fine-grained class differences, and improves the accuracy of small sample image classification, especially with more stable performance under 1-shot settings.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122176362A_ABST
    Figure CN122176362A_ABST
Patent Text Reader

Abstract

The application discloses a small sample image classification method based on a reflection type semantic prompt mechanism and multi-branch semantic alignment, and comprises the following steps: acquiring a support image set corresponding to each category and a category name; performing multi-angle semantic modeling and reflection type compression on the category through the reflection type semantic prompt mechanism, generating and obtaining unified category semantic representation; performing feature extraction on the support image and the image to be classified, obtaining corresponding visual feature representation; inputting the category semantic representation into a multi-branch semantic alignment module, aligning and fusing with the visual feature, obtaining semantic visual fusion feature; calculating the similarity between the category prototype formed based on the support set and the image feature to be classified; if the specified number of training rounds is reached, the training is ended, otherwise the training is continued. The application improves the accuracy of small sample image classification by constructing stable and discriminative category semantic representation and realizing effective fusion of semantic and visual features.
Need to check novelty before this filing date? Find Prior Art