The presence of clutter in ground penetrating radar (GPR) images obscures or distorts the responses of subsurface targets, significantly impacting the accuracy of detecting and identifying targets. Current clutter removal approaches often result in residual clutter or deformation of target responses, especially when confronted with intricate and uneven clutter in real-world GPR signatures. To address this issue in clutter removal under realistic conditions, this research introduces the Spatial Attention-based Convoluted Gazelle Neural Network (SACGNN). SACGNN is trained on large-scale dataset designed to reflect real-time conditions and incorporates a residual block into its spatial attention-based architecture. This integration enhances its ability to suppress clutter and restore target reflections on surfaces like homogeneous, rough, water, and grass surfaces. Tuning the SACGNN using the gazelle optimization algorithm excels in clutter removal by offering faster convergence, maintaining a balanced exploration-exploitation trade-off, and enhancing robustness across various data patterns.

In seismic data processing, denoising is one of the essential steps to identifying the earth’s subsurface layer information. The noise present in the seismic data is categorized into two types: random and erratic noise. The random noise is distributed uniformly over the seismic data. The erratic noise attenuation is always challenging due to the unknown distribution of high-amplitude peaks over seismic data. The existing double sparsity dictionary learning (DSDL) method performs with analytical and adaptive transforms; both the transforms include iterative algorithms with K-singular-value decomposition (SVD); it is computationally costly, and the dictionary is initialized with trained data. To address these limitations, we propose a novel method of dictionary learning with double regularization (DLDR) to denoise both random and erratic noise from seismic data. In double regularization, we used with$\ell _{1}$-norm and nuclear norm. The denoised data is applied to the alternating direction method of multipliers (ADMMs) to improve denoising while preserving the signal features from seismic data while reducing the computational cost. We evaluated the performance of the proposed method using signal-to-noise ratio (SNR), mean squared error (MSE), and local similarity map. The numerical results demonstrated that the proposed method resulted in higher SNR, lower MSE, and less signal leakage from seismic data. The method gives precise interpretation from the denoised seismic data.

The knowledge of hidden resources present inside the earth layers is vital for the exploration of petroleum and hydrocarbons. However, the recorded seismic data is noisy and incomplete with missing traces that leads to misinterpretation of the earth layers. In this manuscript, we consider seismic data with Gaussian, non-Gaussian noise distribution, regular and irregular missing traces. We propose a method for simultaneous noise attenuation and reconstruction of the incomplete seismic data with attention based wavelet convolutional neural network (AWUN). The wavelet transform is used as pooling layer and inverse wavelet transform is used for upsampling layers to avoid information loss. The attention module is used to obtain weights for various feature channels with higher weights assigned to the more significant information. In addition, we propose to use hybrid loss function (logcosh + huberloss) to denoise and accurately reconstruct the seismic data. Moreover, the effect of various hyper-parameters in the training process of convolutional neural networks is studied. Further, we tested the performance of proposed method on synthetically generated data and field data examples. The quantitative results demonstrated that our proposed deep learning method has shown improved signal-to-noise ratio (SNR) and mean squared error (MSE) when compared to the existing state-of-the-art methods.

In seismic data processing, denoising is one of the important steps to get the earth subsurface layers’ information accurately. The dictionary learning (DL) method is one of the prominent methods to denoise the seismic data. In the DL method, there are various parameters involved for denoising such as patch size, dictionary size, number of training patches, choice of threshold, sparsity level, computational cost, and number of iterations for DL. In this work, we study each parameter and its effects on seismic denoising in terms of signal-to-noise ratio and mean square error between the true and denoised seismic data. We examined the performance of the DL method on synthetic and field seismic data for various choices of parameters.

In seismic data processing, attenuation of erratic noise is a challenging task due to the unknown noise distribution. Erratic noise consists of high amplitude peaks and conventional sparse transforms based on least-square (LS) approach that is not appropriate for erratic noise attenuation. An alternative approach, where the data with erratic noise are transformed into pseudodata and then denoised based on fast discrete curvelet transform with structure-oriented space-varying median filtering, is performed and achieves better attenuation. However, the fast discrete curvelet transform with a fixed basis lacks the adaptivity for various data patterns of seismic data. Hence, in this article, we propose a double sparsity dictionary learning (DSDL) method which performs denoising and also preserves the original features of seismic data. The proposed method combines the strength of the analytical transform and adaptive transform to attenuate both random noise and erratic noise in data. The performance of proposed DSDL method is studied on synthetic datasets and field datasets. The numerical results demonstrate that the proposed method gives a better signal-to-noise ratio (SNR), a lower mean-squared error, and energy values for the denoised data in comparison to the existing methods.