Feature fusion method and device for semantic segmentation model, and storage medium

By weighting and expanding the encoder features in the semantic segmentation model and fusing them with the decoder features, the problem of semantic gap between encoder and decoder features is solved, and the image segmentation accuracy of the semantic segmentation model is improved.

CN119027769APending Publication Date: 2024-11-26COWA TECHNOLOGY CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411114693.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

There is a semantic gap in the features extracted by the encoder and decoder in the existing semantic segmentation model. A simple fusion strategy will introduce redundant information, affecting the segmentation effect.

Method used

By determining whether the calculation method of the encoder and decoder features is a series connection method. If so, the encoder features are weighted and expanded to reduce the semantic gap, and convolution operations are performed in the semantic segmentation model to fuse the features.

Benefits of technology

Without changing the model structure, the semantic gap during feature fusion is effectively narrowed and the image segmentation accuracy of the semantic segmentation model is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027769A_ABST
    Figure CN119027769A_ABST
Patent Text Reader

Abstract

The invention discloses a feature fusion method and device for a semantic segmentation model and a storage medium, and the method comprises the following steps: firstly, carrying out the judgment whether all modes are in a serial connection mode, and if all modes are in the serial connection mode; carrying out weighted expansion processing on the first feature extracted by the encoder to obtain an expanded first feature; and finally, fusing the expanded first feature obtained in the step S2 and the second feature extracted from the decoder. According to the technical scheme, after the first features extracted from the encoder are subjected to weighted expansion processing, the first features are fused with the second features extracted from the decoder, and then convolution operation is carried out in the semantic segmentation model, so that the method only increases negligible calculation amount under the condition of not changing the structure of the model; therefore, the semantic gap during feature fusion can be effectively reduced, and the segmentation precision of the semantic segmentation model on the image is improved.
Need to check novelty before this filing date? Find Prior Art