Multi-scale cross-fusion attention for few-shot intestinal polyp image semantic segmentation

Yuxiu Kang; Yaling Zhu; Jundi Wang

doi:10.1117/12.3048130

12 November 2024 Multi-scale cross-fusion attention for few-shot intestinal polyp image semantic segmentation

Yuxiu Kang, Yaling Zhu, Jundi Wang

Author Affiliations +

Proceedings Volume 13395, International Conference on Optics, Electronics, and Communication Engineering (OECE 2024) ; 133952X (2024) https://doi.org/10.1117/12.3048130
Event: International Conference on Optics, Electronics, and Communication Engineering, 2024, Wuhan, China

Abstract

The few-shot intestinal polyp image semantic segmentation aims to segment unseen targets in the query image with only a few pixel-by-pixel annotated support images. However, existing few-shot intestinal polyp image semantic segmentation methods mainly mine valuable guidance information from the support branch, ignoring the role played by the task target information in the query branch in improving model performance. In this paper, a few-shot intestinal polyp image semantic segmentation method based on multi-scale cross fusion attention is proposed. Firstly, the pretrained convolutional neural network is used to map the images on both branches into the same feature space, and the multi-scale information on different branches is extracted respectively. Then, cross attention is used to establish the scale fusion between the multi-scale information of the support branch and the query branch, promoting the semantic alignment of features between branches. Finally, the similarity values between the encoding features at each position on the prototype set and the query image are calculated using a parameter-free metric method, and the unseen target area in the query image is segmented according to the similarity value. Extensive experiments on open-source intestinal polyp image dataset demonstrate the superiority of the designed method.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Yuxiu Kang, Yaling Zhu, and Jundi Wang "Multi-scale cross-fusion attention for few-shot intestinal polyp image semantic segmentation", Proc. SPIE 13395, International Conference on Optics, Electronics, and Communication Engineering (OECE 2024) , 133952X (12 November 2024); https://doi.org/10.1117/12.3048130

Access the abstract

PROCEEDINGS
7 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY