Morphological Component Analysis Research Articles

An image processing mechanism for aerial inspection robots to detect submillimeter-width concrete cracks in social infrastructures

Inspection robots for early detection of potential risks in severe environments require a high accuracy like human experts. A fine mechanism is crucial for extracting target components from noisy signals. We have proposed a detection system for submillimeter-width cracks in concrete surfaces of social infrastructures, such as bridges, by using morphological component analysis (MCA), for aerial image-inspection robots. Traditional schemes like PCA have relied on linear decomposition for the separation of target signal and noise components. Recent advancement in signal decomposition focuses on the enhancement of linearity in the separation by introducing a set of nonlinear basis functions to represent the raw signal even when multiple factors are mixed in a nonlinear manner. In this sense, MCA is a core technique to be able to isolate target components to represent submillimeter-width cracks from others. We proposed a proper pre- and post-processing operations to attach MCA, which demonstrated a high accuracy yet coarse and fine image components have to be integrated redundantly. In the present study, we successfully found a simpler mechanism to set the single basis function to extract the target by introducing a new thresholding mechanism. It suggests a high potential of MCA for inspection robots for various purposes.

X-Let's Atom Combinations for Modeling and Denoising of OCT Images by Modified Morphological Component Analysis.

An improved analysis of Optical Coherence Tomography (OCT) images of the retina is of essential importance for the correct diagnosis of retinal abnormalities. Unfortunately, OCT images suffer from noise arising from different sources. In particular, speckle noise caused by the scattering of light waves strongly degrades the quality of OCT image acquisitions. In this paper, we employ a Modified Morphological Component Analysis (MMCA) to provide a new method that separates the image into components that contain different features as texture, piecewise smooth parts, and singularities along curves. Each image component is computed as a sparse representation in a suitable dictionary. To create these dictionaries, we use non-data-adaptive multi-scale ( X -let) transforms which have been shown to be well suitable to extract the special OCT image features. In this way, we reach two goals at once. On the one hand, we achieve strongly improved denoising results by applying adaptive local thresholding techniques separately to each image component. The denoising performance outperforms other state-of-the-art denoising algorithms regarding the PSNR as well as no-reference image quality assessments. On the other hand, we obtain a decomposition of the OCT images in well-interpretable image components that can be exploited for further image processing tasks, such as classification.

Modeling and sparsity-promoting separation of wind turbine noise in common-shot gathers

In land seismic acquisition, the quality of common-shot gathers is severely degraded by wind turbine noise (WTN) when wind turbines are operating continuously in survey areas. The high-amplitude WTN overlaps or even completely submerges the body and surface waves (signals). Through time-space and frequency analysis, three main features of the WTN are observed: (1) it is periodic with nearly constant frequencies over time, (2) it is coherent but exhibits different apparent velocities in space, and (3) it has relatively narrow bands with varying central frequencies. The first feature enables WTN to distort signals from shallow to deep, whereas the latter two features make traditional methods that separate noise and signals based on velocity and frequency differences less effective. To suppress the WTN, we first analyze its formation and propagation mechanism and then develop a WTN simulation model to validate the presented mechanism. Based on our analysis of WTN and signals, we consider common-shot gathers as the linear superpositions of periodic WTN and relatively broadband signals (referred to as low-oscillatory signals). This additive mixture aligns with the feasibility premise of morphological component analysis (MCA). Finally, based on MCA theory, we develop a sparsity-promoting separation method to suppress WTN in common-shot gathers. To implement our separation method, we construct two dictionaries using the tunable Q-factor wavelet transform (TQWT) and the discrete cosine transform (DCT). TQWT and DCT can sparsely represent the oscillating waves (signals) and periodic waves (WTN), respectively. This work contributes to the existing knowledge of WTN separation by modeling the periodicity of WTN and the low-oscillatory behavior of a signal, rather than relying on velocity or frequency differences. Our method is tested on synthetic and field data, and both tests demonstrate its effectiveness in separating WTN and preserving signals.

Synopsis of Osmunda (royal ferns; Osmundaceae): towards reconciliation of genetic and biogeographic patterns with morphologic variation

Abstract We present an overview of the morphology, biogeography, and ecology of the fern genus Osmunda (i.e. without previously included taxa of Claytosmunda, Osmundastrum, and Plenasium), with a focus on the American O. spectabilis Willd. and Old World O. regalis L. While genetic data supports the separation of these taxa, commonly used morphologic characters to distinguish between O. regalis and O. spectabilis (maximum frond size; pinnae sessile vs. stalked; pinnules opposite vs. alternate; general pinnule proportions) are not infallible. We recorded morphometric differences among and between O. regalis and O. spectabilis and correlated these with spatio-environmental gradients. Our work provides an updated taxonomic overview with full synonymy and diagnostic key of Osmunda. We preliminarily recognize six informal subtaxa in O. regalis (‘regalis’, ‘abyssinica’, ‘huegeliana’, ‘longifolia’, ‘transvaalensis’, ‘obtusifolia’) and three in O. spectabilis (‘spectabilis’, ‘palustris’, ‘piresii’) that separate in a morphological traits-based principal component analysis. The morphotaxa appeared geographically structured and their distribution showed significant correlations with elevation, mean annual precipitation, annual mean temperature, and temperature seasonality. The taxonomic rank and validity of the proposed subtaxa has to be tested through a large-scale sampling, e.g. with comparative cp/nr genomic and cytogenetic analyses. Such future analyses may help to ascertain whether the observed morphologic peculiarities are conditioned solely by abiotic factors (which would allow addressing them as forms) or are manifested in the genome.

A fresh perspective on the 3D dynamics of Tycho’s supernova remnant: Ejecta asymmetries in the X-ray band

Context. Even 450 yr after the explosion of the Type Ia SN 1572, the dynamics of the Tycho supernova remnant (Tycho’s SNR) can offer keys to improving our understanding of the explosion mechanism and the interaction of the remnant with the interstellar medium. Aims. To probe the asymmetries and the evolution of the SNR, we tracked the ejecta dynamics using new methods applied to the deep X-ray observations available in the Chandra space telescope archive. Methods. For the line-of-sight (LoS) velocity measurement (Vz), we used the Doppler effect focused on the bright Si line in the 1.6-2.1 keV band. Using the component separation tool called general morphological component analysis (GMCA), we successfully disentangled the red and blueshifted Si ejecta emission. This allowed us to reconstruct a map of the peak energy of the silicon line with a total coverage of the SNR at a 2″ resolution. We then obtained a proxy of the integrated velocity along the LoS. For the proper motions in the plane of the sky (Vxy), we developed a new method, called Poisson optical flow (POF), to measure the displacement of two-dimensional (2D) features between the observations of 2003 and 2009. The result is a field of around 1700 velocity vectors covering the entire SNR. Results. These exhaustive three-dimensional (3D) velocity measurements reveal the complex dynamics of Tycho’s SNR. Our study sheds light on a patchy Vz map, where most regions are dominated by the foreground or the background part of the shell. On a large scale, an asymmetry is seen, with the north being dominantly blueshifted and the south redshifted. The proper-motion vector field, Vxy, highlights different dynamics between the eastern and the western parts of the SNR. The eastern velocity field is more disturbed by external inhomogeneities and the south-east ejecta knot. In particular, a slow-down is observed in the north-east, which could be due to the interaction with higher densities, as seen in other wavelengths. The vector field is also used to backtrace the center of the explosion, which is then compared with potential stellar progenitors in the area. The latest Gaia DR3 parallax measurements exclude most stellar candidates based on their distances, leaving only stars B and E as possible candidates, at respective distances of 2.53−0.20+0.23 kpc and 3.52−1.0+2.0 kpc, which are consistent with the expected distance range of the SNR at 2.5–4 kpc.

Needlet Karhunen–Loève (NKL): a method for cleaning foregrounds from 21 cm intensity maps

ABSTRACT This paper introduces a technique called needlet Karhunen–Loéve (NKL), which cleans both polarized and unpolarized foregrounds from H i intensity maps by applying a Karhunen–Loéve transform on the needlet coefficients. In NKL, one takes advantage of correlations not only along the line of sight, but also between different angular regions, referred to as ‘chunks’. This provides a distinct advantage over many of the standard techniques applied to map space that one finds in the literature, which do not consider such spatial correlations. Moreover, the NKL technique does not require any priors on the nature of the foregrounds, which is important when considering polarized foregrounds. We also introduce a modified version of Generalized Needlet Internal Linear Combination (GNILC), referred to as MGNILC, which incorporates an approximation of the foregrounds to improve performance. The NKL and MGNILC techniques are tested on simulated maps which include polarized foregrounds. Their performance is compared to the GNILC, generalized morphological component analysis, independent component analysis, and principal component analysis techniques. Two separate tests were performed. One at 1.84 < z < 2.55 and the other at 0.31 < z < 0.45. NKL was found to provide the best performance in both tests, providing a factor of 10–50 improvement over GNILC at $k \lt 0.1\, {\rm hMpc^{-1}}$ in the higher redshift case and $k \lt 0.03 \, {\rm hMpc^{-1}}$ in the lower redshift case. However, none of the methods were found to recover the power spectrum satisfactorily at all baryon acoustic oscillations scales.

Quality-Aware Fusion of Multisource Internal and External Fingerprints Under Multisensor Acquisition

Fingerprint is one of the most widely used biometric features. The researches on external fingerprint collected by total internal reflection and the internal fingerprint obtained by optical coherence tomography have been carried out extensively. Studies proved the consistency between them. The external fingerprint is susceptible to the fingertip status, whereas the internal fingerprint has strong anti-interference and anti-spoofing capabilities. They originate from different skin layers and are collected by multiple sensors. Given the corresponding advantages and disadvantages of external and internal fingerprints, their fusion can maximize effective information and improve fingerprint image quality; thus, this fusion is conducive to fingerprint identification. In this study, a fingerprint fusion method based on the quality-aware convolutional-sparsity-based morphological component analysis (CSMCA) is proposed. The proposed method realizes the fusion of multisource and multisensor fingerprints for the first time. Quality indexes, namely, spatial consistency, dryness, and humidity, are selected. Moreover, a simulation-based combination scheme is proposed for the pixel-level quality representation. The quality index is integrated with CSMCA for quality-aware fusion, which retains high-quality components and reduces low-quality areas. The experiments prove the superiority of the proposed method in quality scores and matching performances, indicating that internal fingerprints can amend external fingerprints. The matching experiments with either external or internal fingerprints show that the fused fingerprints can be compatible with the existing fingerprint databases. Our work can provide references and insights into identifying mutilated fingerprints in the future.

Graph laplacian regularization with sparse coding in secure image restoration and representation for Internet of Things

Image restoration (IR) attempts to recreate the original (ideal) scene from a degraded observation. The goal of image restoration is to avert the process of image deterioration that happens during image acquisition and processing. Blurring and noise are two common types of picture capture degradation. When the blurring function is unknown, the image restoration issue is referred to as blind image restoration. Existing methods of image restoration do not provide efficient results due to time complexity, computational complexity, large-scale scaling factor and limited input. Therefore, in order to overcome those problems, an Improved Graph Laplacian Regularization with Sparse Coding by integrating Internet of Things (IoT) is developed in this research. The process of removing de-noised portions of image to texture layer and cartoon layer is carried out by Morphological Component Analysis (MCA). An improved Graph Laplacian regularized method and Simultaneous Sparse Coding with Gaussian Scale Mixture (SSC-GSM) methods is performed on texture layer, cartoon layer and restored image. A Levin’s dataset and a real-time dataset such as industrial manufacturing products are analyzed in this research. The proposed Graph Laplacian algorithm and sparse coding model are better efficient in identifying sharper texture information and getting a restored image. The proposed method proves that it has better performance in terms of Peak Signal-to-Noise Ratio (PSNR) of 35.82 and Structural Similarity Index Measure (SSIM) of 0.94 than the existing methods.

Morphological component analysis under non-convex smoothing penalty framework for gearbox fault diagnosis

The sparse representation methodology has been identified to be a promising tool for gearbox fault diagnosis. The core is how to precisely reconstruct the fault signal from noisy monitoring signals. The non-convex penalty has the ability to induce sparsity more efficiently than convex penalty. However, the introduction of non-convex penalty usually influences the convexity of the model, resulting in the unstable or sub-optimal solution. In this paper, we propose the non-convex smoothing penalty framework (NSPF) and combine it with morphological component analysis (MCA) for gearbox fault diagnosis. The proposed NSPF is a unify penalty construction framework, which contains many classical penalty while a new set of non-convex smoothing penalty functions can be generated. These non-convex penalty can guarantee the convexity of the objective function while enhancing the sparsity, thus the global optimal solution can be acquired. The simulation and engineering experiments validate that the NSPF enjoys more reconstruction precision compared to the existing penalties.

Robust RFI Excision for Pulsar Signals by a Novel Nonlinear M-type Estimator with an Application to Pulsar Timing

Radio frequency interference (RFI) mitigation for pulsar signals is a long perplexing issue in astrophysical measurements. Linear mitigation methods are often criticized for limited RFI excision range and weakness of RFI modeling. Meanwhile, thresholding methods (e.g., the SumThreshold) suffer from empirical factors. In our opinion, the main defect of the current status is the lack of a concise definition to distinguish signals from RFI with the aid of certain techniques, e.g., sparse representation. This point is the root cause of these problems and also forms our motivation. This paper aims to expand the excision range (e.g., the on-pulse and quasiperiodic RFI) and cut down some empirical factors. The main contribution is that we give a definition and derive a widely practicable nonlinear framework for RFI excision. This framework can overcome the susceptibility of the least-square criterion to RFI, and excise almost all types of RFI once and for all. A robust LnCosh criterion based nonlinear maximum likelihood-type (M-type) penalized smoothing estimator is introduced. The novelty is that this estimator is first embedded into the iterative shrinkage-thresholding algorithm (ISTA) and the fast ISTA. Nonlinearity highlights this method. Curvelet sparsity gives satisfying approximation for pulsar signals containing dispersion feature. Finally, useful signal details will be retrieved from the excision residual by a morphological component analysis. This method is applied to the time-frequency signals collected by the Nanshan 26 m Radio Telescope. The numerical experiments can persuasively prove that it has desired application prospects.

Wind turbine clutter mitigation using morphological component analysis with group sparsity

To address the problem that dynamic wind turbine clutter (WTC) significantly degrades the performance of weather radar, a WTC mitigation algorithm using morphological component analysis (MCA) with group sparsity is studied in this paper. The ground clutter is suppressed firstly to reduce the morphological compositions of radar echo. After that, the MCA algorithm is applied and the window used in the short-time Fourier transform (STFT) is optimized to lessen the spectrum leakage of WTC. Finally, the group sparsity structure of WTC in the STFT domain can be utilized to decrease the degrees of freedom in the solution, thus contributing to better estimation performance of weather signals. The effectiveness and feasibility of the proposed method are demonstrated by numerical simulations.

Simultaneous random plus outlier attenuation and deblending based on L1‐norm misfit function

AbstractDeblending technology aims at separating simultaneous source seismic data between adjacent shots by allowing multiple sources to be shot simultaneously. Conventional deblending methods based on sparse inversion assume that the primary source is coherent, and the secondary source is randomized. The L2‐norm minimization constraint can effectively minimize the Gaussian random noise while deblending in the transform domain. Nonetheless, the L2‐norm misfit function is highly sensitive to outliers, negatively influencing the deblending performance. An effective optimization strategy is developed with deblending in pre‐stack seismic data to eliminate outliers and enhance deblending accuracy. For this reason, we introduce the deblending algorithm in the morphological component analysis framework, modify the L2‐norm misfit function to outlier‐robust L1‐norm and provide the corresponding derivation in detail via the alternating direction method of multipliers. Applications to synthetic and field data sets prove the improved robustness and efficiency of our deblending method.

A BCI framework for smart home automation using EEG signal

Since the first recording in 1924, modern developments in technology have enabled human electroencephalogram (EEG) acquisition as a non-invasive process, enabling a multitude of opportunities to learn about human brain dynamics. With the capability to tap into localized brain dynamics, it is possible to correlate event-related potentials such as meditation, concentration, and motor control with localized brain activities, opening up a broad spectrum for exploration and implementation in prosthetic, control, and brain computer interfaces. In this work, an attempt has been made to explore human emotions and other intelligent states that can be interpreted to automate and control electrical appliances for differently abled people. A smart home automation model has been designed and implemented using a Think Gear Application-specific integrated circuit (ASIC) Module (TGAM) EEG sensor module integrated with a Bluetooth module, which serves as the core for control applications. A combination of external triggers and brain states are interpreted and forwarded to gain control of the connected appliances via a local server over the internet. Equipped with internet connectivity and Internet of Things (IoT), the system also facilitates long-distance communication and control, which can be easily translated to industrial control and automation applications. Based on a single-channel analog EEG signal acquisition module, this study details the development of a Brain Computer Interface (BCI) control and monitoring system for smart home automation with blink and attention features. The designed BCI system has a large bandwidth of 400 Hz, an easy setup, Morphological Component Analysis (MCA) based blink detection, monitoring, control of devices, and high accuracy at a low cost. Each subject completed three trials separated by one minute. The smart devices underwent testing in two states, namely on and off. The response time and accuracy of the system were recorded for each trial. The average system response time for the devices was determined to be 17.13 sec for switching ON and 20.66 sec for switching OFF, with an accuracy of 98.73% and 95.75% for ON and OFF states respectively.

Approximate extraction of late-time returns via morphological component analysis

A fundamental challenge in acoustic data processing is to separate a measured time series into relevant phenomenological components. A given measurement is typically assumed to be an additive mixture of myriad signals plus noise whose separation forms an ill-posed inverse problem. In the setting of sensing elastic objects using active sonar, we wish to separate the early-time return from the object's geometry from late-time returns caused by elastic or compressional wave coupling. Under the framework of morphological component analysis (MCA), we compare two separation models using the short-duration and long-duration responses as a proxy for early-time and late-time returns. Results are computed for a broadside response using Stanton's elastic cylinder model as well as on experimental data taken from an in-air circular synthetic aperture sonar system, whose separated time series are formed into imagery. We find that MCA can be used to separate early and late-time responses in both the analytic and experimental cases without the use of time-gating. The separation process is demonstrated to be compatible with image reconstruction. The best separation results are obtained with a flexible, but computationally intensive, frame based signal model, while a faster Fourier transform based method is shown to have competitive performance.

Data-driven morphological component analysis

A common goal in signal processing is to decompose a signal into constituent components. Morphological Component Analysis (MCA) is a convex optimization technique that can be used to decompose a signal into a sum of sparse representations determined by chosen dictionaries. Previously, MCA has been shown to successfully decompose sonar data into long-duration and short duration components, using Enveloped Sinusoid Parseval (ESP) frames as the dictionaries. However, in its current state, ESP MCA is model-driven and requires prior selection of envelope parameters. This presentation describes the addition of hyper-parameter tuning to ESP MCA. Specifically, gradient descent is performed on an unrolled version of the ESP MCA regularization algorithm in order to learn the optimal envelope parameters used to construct the ESP frames. Addition of hyper-parameter tuning allows the dictionary atoms to adapt and thus improves sparse representation of the sensed data.

Approximate extraction of late-time returns via morphological component analysis

A fundamental challenge in acoustic data processing is to separate a measured time series into relevant phenomenological components. In the setting of sensing elastic objects using active sonar, we wish to separate the early-time returns (e.g., returns from the object's exterior geometry) from late-time returns caused by elastic or compressional wave coupling. Under the framework of morphological component analysis (MCA), we compare two separation models using the short-duration and long-duration responses as a proxy for early-time and late-time returns. Results are computed for broadside geometries using Stanton's elastic cylinder model as well as experimental data taken from an in-air circular synthetic aperture sonar (AirSAS) system, whose separated time series are formed into imagery. We find that MCA can be used to separate early and late-time responses in both the analytic and experimental cases without the use of time-gating. The separation process is demonstrated to be robust to noise and compatible with AirSAS image reconstruction. The best separation results are obtained with a flexible, but computationally intensive, frame based signal model, while a faster Fourier Transform based method is shown to have competitive performance.

Self-adaptive seismic data reconstruction and denoising using dictionary learning based on morphological component analysis

Data reconstruction and data denoising are two critical preliminary steps in seismic data processing. Compressed Sensing states that a signal can be recovered by a series of solving algorithms if it is sparse in a transform domain, and has been well applied in the field of reconstruction, when, sparse representation of seismic data is the key point. Considering the complexity and diversity of seismic data, a single mathematical transformation will lead to incomplete sparse expression and bad restoration effects. Morphological Component Analysis (MCA) decomposes a signal into several components with outstanding morphological features to approximate the complex internal data structure. However, the representation ability of combined dictionaries is constrained by the number of dictionaries, and cannot be self-adaptively matched with the data features. Dictionary learning overcomes the limitation of fixed base function by training dictionaries that are fully suitable for processed data, but requires huge amount of time and considerable hardware cost. To solve the above problems, a new dictionary library (K-Singluar Value Decomposition learning dictionary and Discrete Cosine Transform dictionary) is hereby proposed based on the efficiency of fixed base dictionary and the high precision of learning dictionary. The self-adaptive sparse representation is achieved under the Morphological Component Analysis framework and is successfully applied to the reconstruction and denoising of seismic data. Real data tests have proved that the proposed method performs better than single mathematical transformation and other combined dictionaries.

IM-EDRD from Retinal Fundus Images Using Multi-Level Classification Techniques

In recent years, there has been a significant increase in the number of people suffering from eye illnesses, which should be treated as soon as possible in order to avoid blindness. Retinal Fundus images are employed for this purpose, as well as for analysing eye abnormalities and diagnosing eye illnesses. Exudates can be recognised as bright lesions in fundus pictures, which can be the first indicator of diabetic retinopathy. With that in mind, the purpose of this work is to create an Integrated Model for Exudate and Diabetic Retinopathy Diagnosis (IM-EDRD) with multi-level classifications. The model uses Support Vector Machine (SVM)-based classification to separate normal and abnormal fundus images at the first level. The input pictures for SVM are pre-processed with Green Channel Extraction and the retrieved features are based on Gray Level Co-occurrence Matrix (GLCM). Furthermore, the presence of Exudate and Diabetic Retinopathy (DR) in fundus images is detected using the Adaptive Neuro Fuzzy Inference System (ANFIS) classifier at the second level of classification. Exudate detection, blood vessel extraction, and Optic Disc (OD) detection are all processed to achieve suitable results. Furthermore, the second level processing comprises Morphological Component Analysis (MCA) based image enhancement and object segmentation processes, as well as feature extraction for training the ANFIS classifier, to reliably diagnose DR. Furthermore, the findings reveal that the proposed model surpasses existing models in terms of accuracy, time efficiency, and precision rate with the lowest possible error rate.

Robust sparse time‐frequency analysis for data missing scenarios

Sparse time-frequency analysis (STFA) can precisely achieve the spectrum of the local truncated signal. However, when the signal is disturbed by unexpected data loss, STFA cannot distinguish effective signals from missing data interferences. To address this issue and establish a robust STFA model for time-frequency analysis (TFA) in data loss scenarios, a stationary Framelet transform-based morphological component analysis is introduced in the STFA. In the proposed model, the processed signal is regarded as a sum of the cartoon, texture and data-missing parts. The cartoon and texture parts are reconstructed independently by taking advantage of the stationary Framelet transform. Then, the signal is reconstructed for STFA. The forward-backwards splitting method is employed to split the robust STFA model into the data recovery and robust time-frequency imaging stages. The two stages are then solved separately by using the alternating direction method of multipliers (ADMM). Finally, several experiments are conducted to show the performance of the proposed robust STFA method under different data loss levels, and it is compared with some existing state-of-the-art time-frequency methods. The results indicate that the proposed method outperforms the compared methods in obtaining the sparse spectrum of the effective signal when data are missing. The proposed method has a potential value in TFA in scenarios where data is easily lost.

Improving sparse representation with deep learning: A workflow for separating strong background interference

Revealing hidden reservoirs that are severely shielded by strong background interference (SBI) is critical to subsequent refined interpretation. To enhance the characterization of these reservoirs, current interpretation workflows merge multiple attribute information, necessitating intensive human expertise. As an alternative, we regard SBI suppression as a signal separation problem and develop a workflow to suppress SBI by cascading a sparse representation method and deep learning. SBI has coherent morphological characteristics in seismic sections; reservoir seismic responses, such as channels and karst caves, have a narrow spatial distribution, exhibiting abrupt morphological characteristics. As their morphologies differ, we select two 2D sparse representation dictionaries to identify their individual components. Through the morphological component analysis (MCA) technique, we can obtain adequate SBI separation results. However, the MCA separation is inevitably limited because 2D dictionaries cannot adequately represent 3D structures, but 3D dictionaries are not viable due to computing constraints. As an extension, we use 3D deep learning to improve the separation results based on the 2D MCA results. Specifically, the network is fed with training samples from a region with better SBI suppression results obtained by the MCA method. After learning a direct mapping from noisy data to SBI, the network can improve the separation results and remove more SBI than the previous conventional method. Field data experiments demonstrate that our separation workflow successfully enhances reservoir structures after removing SBI.