| 摘要: |
| 为了提高频谱利用率，提出了一种基于联邦学习（Federated Learning,FL）、Transformer和卷积神经网络（Convolutional Neural Network,CNN）的水声智能频谱感知算法。首先，基于FL实现数据隔离状态下的信息共享，并应用Paillier加密技术进行权重加密保障；其次，本地感知数据经连续小波变换构建为时频谱图;最后，融合CNN与Transformer构建了TransCNN感知器，通过并行分支实现了高精度感知。在信噪比-18獈0 dB范围内，与RepVGG、Swin-Transformer、YOLOv7、MobileNet算法相比，所提的水声智能频谱感知算法的平均检测概率提升了4%~10%，平均虚警概率降低了2%~9%。 |
| 关键词: 海洋物联网 智能频谱感知 联邦学习 连续小波变换 深度可分离卷积 |
| DOI：10.20079/j.issn.1001-893x.240911001 |
|
| 基金项目:国家自然科学基金资助项目(62201313)；数字化学习技术集成与应用教育部工程研究中心创新基金项目(1321012) |
|
| An Underwater Acoustic Intelligent Spectrum Sensing Algorithm Based on FL-TransCNN Neural Network |
| LI Yufang,WANG Kai,ZHANG Liliang,XU Lingwei,Thomas Aaron Gulliver |
| (1.College of Information Science & Technology,Qingdao University of Science & Technology,Qingdao 266061,China;2.Engineering Research Center of Integration and Application of Digital Learning Technology,Ministry of Education,Beijing 100039,China;3.Department of Electrical and Computer Engineering,University of Victoria,Victoria V8W 3P6,Canada) |
| Abstract: |
| To improve the spectrum utilization,an underwater acoustic intelligent spectrum sensing algorithm based on federated learning(FL),Transformer and convolutional neural network(CNN) is proposed.Firstly,information sharing in a data isolation state is realized based on FL,and Paillier encryption technology is applied to guarantee weight encryption.Secondly,the local sensing data is constructed into a time-frequency spectrum by continuous wavelet transform.Finally,a TransCNN perceptron is constructed by combining CNN and Transformer,and high-precision perception is achieved through parallel branches.Compared with that of RepVGG,Swin-Transformer,YOLOv7,and MobileNet algorithms,the average detection probability of the proposed algorithm based on the FL-TransCNN neural network is improved by 4% to 10% and the average false alarm probability is reduced by 2% to 9% in -18 dB to 0 dB signal-to-noise ratio. |
| Key words: marine Internet of Things intelligent spectrum sensing federated learning continuous wavelet transform depthwise separable convolution |
《电讯技术》编辑部 