LEARNING FROM NOISY DATA AN UNSUPERVISED RANDOM DENOISING METHOD FOR SEISMIC DATA USING MODEL-BASED DEEP LEARNING
ABSTARCT :
For the noise removal problem of noisy seismic data, an improved noise reduction technique based on feedforward denoising neural network (DnCNN) is proposed. The previous DnCNN, which was designed to minimise noise in seismic data, had an issue with a large network depth, which hampered training efficiency. The revised DnCNN technique was previously introduced for noise reduction in natural data sets, and after modifying the essential parameters, this study extends the algorithm to noise reduction in seismic dataThe DUDnCNN algorithm can reduce noise with high efficiency, according to the analysis and comparison of the experimental findings, and the method has certain feasibility and significance for further seismic data noise reduction research.
EXISTING SYSTEM :
A non-diagonal seismic noise reduction technique based on continuous wavelet transform and hybrid block thresholding was proposed by Mousavi et al. The full-variance regularised nonlocal mean approach was used by Li et ato reduce noise in seismic data, which efficiently removed the noise while maintaining the edges. A noise reduction technique based on dynamic clustering with singular value decomposition was suggested by Wang Wei et al[10]. The adaptive thresholding approach based on the shearlet transform was used to reduce seismic data noise by Cheng Hao et al . Although many of the aforementioned algorithms have shown promise in the processing of seismic data noise, further research is needed to increase the efficiency and accuracy of noise reduction
DISADVANTAGE :
Because it requires the same distribution of test and training data, the former ones have a problem with poor generalisation ability. We present a novel deep learning framework for attenuating random noise in prestack seismic data in an unsupervised manner to avoid constructing noise-free labels. . And the DUDnCNN loss change curve shows that the training loss has basically converged at 20 iterations, and the peak signal-to-noise ratio has also improved significantly before and after iteration, but even though we try to alleviate the problem of effective information being removed by adjusting the learning rate and Batch size several times, the problem of effective information being removed persists.
PROPOSED SYSTEM :
? In contrast to existing DnCNN and other AI-based denoisers, FFCNN has several appealing features: 1) downsampling and upscaling operations, which can significantly reduce runtimes and memory requirements while maintaining denoising performance, and 2) we introduced noised level maps, which can allow a single convolutional neural network (CNN) model to handle noise models with varying parameters. The main work and benefits of this article are focused on the following two components forreal seismic data denoised work
ADVANTAGE :
? The suggested CNN denoising model was tested with synthetic and field data. When compared to two commonly used state-of-the-art denoising approaches, the experimental results show that random noise attenuation while keeping amplitude is more successful. To get around it, several researchers used labels made from realistic-looking synthetic data or denoised results obtained using traditional approaches. . The adaptive thresholding approach based on the shearlet transform was used to reduce seismic data noise by Cheng Hao et
|