ISMU-Net: insulator image segmentation with mamba-based U-Net

Xunxing Liu

doi:10.1117/12.3049373

12 November 2024 ISMU-Net: insulator image segmentation with mamba-based U-Net

Xunxing Liu

Author Affiliations +

Proceedings Volume 13395, International Conference on Optics, Electronics, and Communication Engineering (OECE 2024) ; 133952K (2024) https://doi.org/10.1117/12.3049373
Event: International Conference on Optics, Electronics, and Communication Engineering, 2024, Wuhan, China

Abstract

Accurate segmentation of insulators in power systems is crucial for localizing and detecting insulator defects, ensuring the safe and efficient transmission of electricity. However, current insulator segmentation models suffer from high misclassification rates and low segmentation accuracy when segmenting aerial insulator images. To address this issue and achieve precise segmentation of insulator images, we propose an Insulator image Segmentation with Mamba-based U-Net (ISMU-Net). Firstly, an enhanced feature extraction block, grounded in visual state modeling, is devised to replace the convolutional blocks of U-Net, enabling comprehensive extraction of insulator image information. Secondly, to mitigate information loss at skip connections and harness the underlying network information, we integrate the attention mechanism from SE-Net into the feature extraction block, optimizing feature fusion at skip connections. Experimental results on a collected dataset of aerial insulator images reveal that ISMU-Net achieves a Precision (Pre) of 92.7%, Recall (Rec) of 91.7%, and F-Measure (F1) of 94.8%. Moreover, ISMU-Net demonstrates strong generalization capabilities across diverse backgrounds, thereby validating its effectiveness in enhancing the accuracy and robustness of insulator segmentation.

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

Citation Download Citation

Xunxing Liu "ISMU-Net: insulator image segmentation with mamba-based U-Net", Proc. SPIE 13395, International Conference on Optics, Electronics, and Communication Engineering (OECE 2024) , 133952K (12 November 2024); https://doi.org/10.1117/12.3049373

Access the abstract

PROCEEDINGS
8 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Image segmentation

Feature extraction

Image processing

Education and training

Inspection

Visualization

Data modeling

Show All Keywords

RELATED CONTENT

Normalized determinant pooling layer in CNNs for multi-label classification
Proceedings of SPIE (July 19 2023)

Research on rock image classification algorithm based on multi task...
Proceedings of SPIE (October 16 2023)

Fine grained image recognition based on multi branch and multi...
Proceedings of SPIE (August 01 2023)

Bias-variance knowledge distillation based on auxiliary classifier
Proceedings of SPIE (October 16 2024)

A visual segmentation method for off-line handwritten Uyghur text
Proceedings of SPIE (June 13 2024)

Utilizing deep convolutional neural networks and transfer learning for bird...
Proceedings of SPIE (August 07 2024)

Subscribe to Digital Library

Receive Erratum Email Alert

Show All Keywords

Keywords/Phrases

Search In:

Publication Years