Research On Seismic Denoising Based On The Sparse Representation And Dictionary Learning

The elasticity and density are different because of the differences of underground structures and medium. Seismic exploration is a kind of geophysical exploration methods that observes and analyzes the propagative rule of the seismic wave underground to explore the oil and gas energy. In seismic exploration, the noise severely distorts and interferes with seismic signal, consequently, the denoising process is very important. The paper is based on the theory of sparse representation of seismic denoising method, we used to study the sparse representation denoising method and flow using fixed dictionary and learning dictionary. In the sparse representation based on the understanding of the principle of the noise reduction algorithm, we propose a new denoising method st K-SVD, we can conclude that this method can effectively suppress the random noise and improve the signal-to-noise ratio, the result is satisfactory.In seismic records, there are so many kinds of noises that varies denoising methods have been well developed. The paper firstly studies traditional seismic denoising methods and conducts a series of denoising experiments including F–X deconvolution filter, wavelet threshold denoising and curvelet threshold denoising. It is can be seen from the comparison of the results: the result of F–X deconvolution filter losses the most effective signals. The wavelet transform is not be able to express the seismic signals along the edges. The method based on the curvelet transform is multi-directional and multi-scale and it can capture the signal’s characteristics optimally. The effective signals is well protected after denosing process.We secondly analyze overcomplete dictionary denoising method based on the theory of sparse representation, and study a fixed DCT dictionary, MOD dictionary and a learning type overcomplete dictionary based on K-SVD algorithm. From the experiment result, we could draw a conclusion that, comparing with the fixed dictionary, the learning type dictionary can gain information including inherent characteristic of data visa constant learning and training, so this method possesses better denoising effect. However, we continued comparing the two learning type overcomplete dictionary, finally, we find that the learning type overcomplete dictionary based on K-SVD algorithm can describe the essential trait of seismic data, it can represent seismic data more sparsely and gain better denoising result and improve the signal-to-noise ratio. Because this method exploits singular value decomposition when it updates dictionary.In the paper, we propose a new algorithm St K-SVD, which exchange STOMP for OMP in K-SVD algorithm. After experiment, we can see that the new method can remove more noise and further improve the signal-to-noise ratio than K-SVD algorithm.