Memory-assisted seismic signal compression based on dictionary learning and sparse coding

Seismic traces recorded in a single sensor from multiple shots demonstrate significant correlation. We propose a memory-assisted seismic signal compression method based on dictionary learning and sparse coding that would explore this correlation. Different from traditional methods, the dictionary used for compression is learned and updated by the information extracted from the common memory between the sender (sensor) node and the receiver node, over a fixed window of the most recent traces. The common memory is formed by the previously transmitted traces. This reduces the cost of communication significantly, a key property for wireless gathering in seismic acquisitions. In particular, seismic traces recorded in a single sensor are segmented into several patches for compression. To utilize the correlation of both intra-patch and inter-patches, an objective function is introduced. Then a memory-assisted dictionary learning algorithm is proposed to optimize and generate the desired dictionary for sparse coding. Experimental results demonstrate that our method outperforms other alternative techniques. Compared to the offline method (without memory), the average compression gain is more than 30%.