Repository files navigation

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

To address the exponentialincrease in seismic data, a variety of methods for seismic data compression havebeen created. In this work, we explore some of the different methods of seismicdata compression. In this project, convolutionalautoencoder models, discrete cosine transform (DCT), and discrete wavelettransform (DWT) models are implemented. Further, quantization techniques is used with the autoencoder model to create amodel that gives much higher compression ratios as compared to the rest. All themodels are compared on the Utah FORGE dataset and are quantitatively analyzedusing 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