Weak Labeling-specific Emitter Identification Algorithm Based on the Weakly Supervised Wav-KAN Network

Weak Labeling-specific Emitter Identification Algorithm Based on the Weakly Supervised Wav-KAN Network

Most existing specific emitter identification technologies rely on supervised learning, making them unsuitable for scenarios with label loss due to factors such as the acquisition environment (e.g., weather conditions, terrain, obstacles, and interference sources), device performance (e.g., radar re...

Full description

Saved in:
Bibliographic Details
Main Authors: Kangsheng LIU, Qing LING, Wenjun YAN, Limin ZHANG, Keyuan YU, Hengyan LIU
Format: Article
Language:English
Published: China Science Publishing & Media Ltd. (CSPM) 2025-04-01
Series:Leida xuebao
Subjects:
specific emitter identification (sei)
weakly supervised wav-kan (wsw-kan)
pseudo-label iteration
weakly supervised learning
contrastive learning
neural network
Online Access:https://radars.ac.cn/cn/article/doi/10.12000/JR24248
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Most existing specific emitter identification technologies rely on supervised learning, making them unsuitable for scenarios with label loss due to factors such as the acquisition environment (e.g., weather conditions, terrain, obstacles, and interference sources), device performance (e.g., radar resolution, signal processing capabilities, and hardware failures), and tagger level. In this study, a weakly labeled specific emitter identification algorithm based on the Weakly Supervised Wav-KAN (WSW-KAN) network is proposed. First, a WSW-KAN baseline network is constructed by integrating the unique learnable edge function of the KAN network with the multiresolution analysis of the wavelet function. The weakly labeled dataset is then divided into a small labeled dataset and a large unlabeled dataset, with the small labeled dataset used for initial model training. Finally, based on the pretrained model, Adaptive Pseudo-Label Weighted Selection (APLWS) is used to extract features from the unlabeled data using a contrast learning method, followed by iterative training, thereby effectively improving the generalization capability of the model. Experimental validation using a real acquisition radar dataset demonstrates that the proposed algorithm achieves a recognition accuracy of approximately 95% for specific emitters while maintaining high efficiency, a small parameter scale, and strong adaptability, making it suitable for practical applications.
ISSN:2095-283X

Similar Items