Implementation of Autoencoder, DCT and DWT models for Seismic Data Compression.

To address the exponential increase in seismic data, a variety of methods for seismic data compression have been created. In this work, we explore some of the different methods of seismic data compression. In this project, convolutional autoencoder models, discrete cosine transform (DCT), and discrete wavelet transform (DWT) models are implemented. Further, quantization techniques is used with the autoencoder model to create a model that gives much higher compression ratios as compared to the rest. All the models are compared on the Utah FORGE dataset and are quantitatively analyzed using the NMSE (Normalised Mean Square Error), NRMSE (Normalised Root Mean Square Error) and SNR (Signal to Noise Ratio) metrics.

This project was done for Information Processing and Compression Course from Sep-Dec 2021.

Project Members:

Roshan Rangarajan

Rohan Jijju