A Generalized Convolutional Autoencoder Framework for Seismic Waveform Compression

Authors

Emad B. Helal, Seismology Department, National Research Institute of Astronomy and Geophysics (NRIAG), Helwan 11421, Egypt AND Cyber Security Department, College of Engineering, Almaaqal University, Basrah 61014, IraqFollow
Mostafa M. Ibrahim, Electrical Engineering Department, Faculty of Engineering, Minia University, Minia 61519, Egypt AND Communications and Computers Engineering Department, Faculty of Engineering, Nahda University in Beni Suef, Beni Suef 62764, Egypt
Ali G. Hafez, Seismology Department, National Research Institute of Astronomy and Geophysics (NRIAG), Helwan 11421, Egypt AND Control and Computer Department, College of Engineering, Almaaqal University, Basrah 61014, Iraq

Abstract

Effective data compression methods have become crucial as the amounts of seismic data are growing exponentially. Due to extensive surveys and ongoing real-time monitoring, compression not only reduces storage requirements but also optimizes communication bandwidth between remote seismic stations and central processing hubs. In this paper, a generalized autoencoder-based compression framework is proposed.. Its primary innovation is a single, fixed network architecture that delivers robust performance across low, medium, and high compression ratios (CRs). Performance was compared with a dynamic pooling-based model and three wavelet-based methods (db4, sym8, coif3) across CRs from 2 to 100. Experimental results demonstrate that the proposed generalized autoencoder establishes a new benchmark for seismic compression. While wavelet methods excel at low compression ratios, our model demonstrates clear superiority in high-compression and real-time scenarios. It robustly outperforms a dynamic autoencoder across the board and dominates at low and mid CRs. The proposed model achieves a superior SNR of 15.80 dB versus wavelet_db4’s 2.92 dB at CR = 100. Combined with a 2–3x speed advantage over wavelet techniques. These results indicate that the generalized approach provides a more reliable balance between compression efficiency and signal quality.

Recommended Citation

Helal, Emad B.; Ibrahim, Mostafa M.; and Hafez, Ali G. (2025) "A Generalized Convolutional Autoencoder Framework for Seismic Waveform Compression," Almaaqal Journal of Sustainability and Emerging Technology: Vol. 1: Iss. 1, Article 6.
Available at: https://ajset.almaaqal.edu.iq/journal/vol1/iss1/6