基于生成对抗网络的超声数据压缩方法研究
Research on Ultrasonic Data Compression Method Based on Generating Countermeasure Network
李泽宇 1王黎明 1聂鹏飞 1韩星程 1武国强 1马文1
作者信息
- 1. 中北大学信息与通信工程学院 太原 030051
- 折叠
摘要
近年来中国工业超声检测发展迅速,几乎涵盖了所有工业领域,然而随着超声检测的不断发展,超声检测的数据量也变得越来越大,这对于数据的传输和存储造成了一定的困扰.针对以上问题论文利用卷积神经网络,长短期记忆网络和生成对抗网络相结合的一种新的网络框架,对超声数据进行特征提取、编码、传输,在生成器中完成对压缩数据的重建,实现比传统超声数据压缩方法更高的压缩比率,同时保证高的还原度.从而减轻数据传输和存储过程中的负载.仿真结果表明,该方法能过实现比传统压缩方法具有更低的压缩率,同时保证较高的还原度.
Abstract
In recent years,industrial ultrasonic testing in China has developed rapidly,covering almost all industrial fields.However,with the continuous development of ultrasonic testing,the amount of data in ultrasonic testing has also become increasing-ly large,which has caused certain difficulties for data transmission and storage.In response to the above issues,this paper uses a new network framework that combines convolutional neural networks,short-term memory networks,and generating confrontation networks to extract,encode,and transmit ultrasonic data.At the receiver,generating confrontation networks are used to reconstruct compressed data,achieving a higher compression ratio than traditional ultrasonic data compression methods,while ensuring a high degree of restoration.Thereby reducing the load during data transmission and storage.Simulation results show that this method can achieve a lower compression rate than traditional compression methods,while ensuring a higher degree of restoration.
关键词
深度学习/数据压缩/超声数据Key words
deep learning/data compression/ultrasonic data引用本文复制引用
基金项目
国家自然科学基金青年基金(62203405)
山西省重点研发计划(2022ZDYF079)
山西省应用基础研究计划(20210302124545)
出版年
2024
国家科技期刊平台
NSTL
万方数据