Desert seismic random noise reduction framework based on improved PSO-SVM

As one of the major regions of carbonate rock oil-gas exploration in western China, Tazhong area of the Tarim Basin has severe environment and complex ground surface conditions, hence the signal to noise ratio (SNR) of the field seismic data is extremely low. To improve the SNR of desert seismic data is a crucial step in the following work. However, the random noise in desert seismic characterizes by non-stationary, non-gaussian, non-linear and low frequency, which are very different from the random Gaussian noise. In addition, the effective signals of desert seismic generally share the same frequency band with strong random noise. These all make some traditional denoising methods cannot suppress it well. Therefore, a new noise suppression framework based on improved PSO-SVM is proposed in this paper. First, we extract the correlation of noisy desert seismic data to form feature vector. Subsequently, the model of improved PSO-SVM was built to classify the extracted feature, thereby identifying the position of the seismic events. Finally, second-order TGV filter was applied for obtaining denoised results. We perform tests on synthetic and field desert seismic record and the denoising results show that the proposed method can effectively preserve effective signals and eliminate random noise.