Kans-Unet Model and Its Application in Image Patch-Shaped Detection

This study proposes a new detection method Kans-Unet, which combines Kans and U-net architecture. Specifically, the method embeds a convolution module based on Kans (Kan-Conv) in the encoder. The module applies a learnable activation function at the edge of the network, which not only reduces the nu...

Full description

Saved in:
Bibliographic Details
Main Authors: Xingsu Li, Zhong Li, Jianping Huang, Ying Han, Kexin Zhu, Bo Hao, Junjie Song, Yumeng Huo
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10969762/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study proposes a new detection method Kans-Unet, which combines Kans and U-net architecture. Specifically, the method embeds a convolution module based on Kans (Kan-Conv) in the encoder. The module applies a learnable activation function at the edge of the network, which not only reduces the number of model parameters but also significantly improves the generalization performance of the network. Secondly, in the transition region between encoder and decoder, the Feature Pyramid Attention (FPA) module is introduced to enhance the ability of the model to capture and analyze multiple scale features. In the decoder part, the CBAM attention mechanism module is integrated to effectively enhance the network’s attention to key information, and to improve the expression ability of image features. It is found that a wide range of patch-shaped anomaly regions are distributed in the VLF band power spectrum image of the space ionospheric electric field, and these patch-shaped anomaly regions are effectively detected by using the Kans-Unet model. The experimental results show that the KANs-Unet algorithm exhibits better detection performance compared to current mainstream semantic segmentation algorithms in the task of detecting abnormal speckle areas. It solves the problem of long model training time caused by insufficient computing power and provides a new method for the detection and analysis of abnormal features in power spectrum images.
ISSN:2169-3536