Noise Corruption Research Articles

Tensor Decomposition-Inspired Convolutional Autoencoders for Hyperspectral Anomaly Detection

Anomaly detection from hyperspectral images (HSI) is an important task in the remote sensing domain. Considering the three-order characteristics of HSI, many tensor decomposition based hyperspectral anomaly detection (HAD) models have been proposed and drawn much attention during the past decades. However, as most tensor decomposition based detectors are directly performed on the original HSI, the detection accuracy is usually limited due to the high-dimension and noise corruption of the HSI. Benefiting from the good capacity of autoencoders (AE) for feature extraction, in this paper, an enhanced tensor decomposition-inspired convolutional AE for HAD is proposed to address those problems, named TDNet. Within the proposed TDNet, the traditional canonical-polyadic (CP) tensor decomposition model is innovatively alternated by a deep neural network (DNN), and the DNN tensor decomposition model performs more stably and robustly for noise. Specificly, a potential abnormal pixels remove strategy is firstly built to obtain the background training sets. Then, a DNN tensor decomposition-inspired convolutional AE is used to recover the original background information, which consists of an encoder, a low-rank tensor decomposition network, and a decoder. Finally, the residual errors between input HSI and recovered background are used for anomaly detection. Extensive experiments demonstrate the superiority of the TDNet in terms of both AUC values and ROC curves.

Nearly Exact Discrepancy Principle for Low-Count Poisson Image Restoration.

The effectiveness of variational methods for restoring images corrupted by Poisson noise strongly depends on the suitable selection of the regularization parameter balancing the effect of the regulation term(s) and the generalized Kullback–Liebler divergence data term. One of the approaches still commonly used today for choosing the parameter is the discrepancy principle proposed by Zanella et al. in a seminal work. It relies on imposing a value of the data term approximately equal to its expected value and works well for mid- and high-count Poisson noise corruptions. However, the series truncation approximation used in the theoretical derivation of the expected value leads to poor performance for low-count Poisson noise. In this paper, we highlight the theoretical limits of the approach and then propose a nearly exact version of it based on Monte Carlo simulation and weighted least-square fitting. Several numerical experiments are presented, proving beyond doubt that in the low-count Poisson regime, the proposed modified, nearly exact discrepancy principle performs far better than the original, approximated one by Zanella et al., whereas it works similarly or slightly better in the mid- and high-count regimes.

Hierarchical soft measurement of load current and state of charge for future smart lithium-ion batteries

Accurate current measurement is indispensable for the management of lithium-ion battery (LIB), especially for the state-of-charge (SOC) estimation. However, accurate current sensing is challenging in electric vehicles (EVs) due to the electromagnetic interference. Moreover, the currents across the parallel branches of battery pack are even unmeasurable due to the absence of current sensor. Motivated by this, this paper proposes a hierarchical soft measurement framework for the load current and SOC addressing different degrees of current sensor uncertainty. Rooted from a common least squares (LS)-based state optimization problem, a total least square (TLS)-based modification is proposed and solved to compensate for the measurement disturbances, and in accordance to estimate the SOC more accurately. One step further, an input-free optimization method is proposed to co-estimate the SOC and load current without using the current measurements. Simulation and experimental results suggest that the proposed hierarchical framework can realize high-fidelity co-estimation of the SOC and load current, especially in the adverse scenarios of both strong noise corruption and current sensor malfunction/missing. The encouraging results open new paradigms for both the high-robustness current-free SOC estimation and the hardware-free soft current measurement of LIB.

DeepBHCP: Deep neural network algorithm for solving backward heat conduction problems

This paper extends a deep neural network method, a semi-supervised one, to solve backward heat conduction problems which have been long-standing computational challenges due to being ill-posed. The effectiveness and robustness of the methodology are demonstrated through various problems including different types of boundary conditions, several types of time-dependent thermal diffusivity factors, and a variety of domains for two-dimensional tests. In spite of traditional methods, there is no need to apply any regularization technique. According to simulation results, this revolutionary strategy can efficiently and accurately extract the pattern of the solutions even when the noise corruption up to ten percent is imposed on input data. Moreover, when the final time is increased further, this approach is efficient in recovering the data at the initial time, which accentuates the method's robustness. To demonstrate the superiority of the semi-supervised neural network process, we make a comparison with the radial basis functions finite difference (RBF-FD) method which is a localized RBF method.

Feature engineering to cope with noisy data in sparse identification

Dynamical systems play a fundamental role related understanding phenomena inherent to several fields of science. Technological advances over the previous several decades have generated a large amount of data that might be used in the inference of dynamical systems. Regardless of the sensor types adopted to perform the data acquisition procedure, it is useful to verify the existence of certain noise corruption in data. Generically, system identification is directly affected by noisy scenarios, which result in the false discovery of non-spurious models. In this work, we demonstrate how the hybridization of several machine learning techniques improves the robustness to noise with respect to the system identification assignment, advancing a pioneer methodology known as Sparse Identification of Nonlinear Dynamics (SINDy). Specifically, in the current work, we show the success of the proposed strategy from numerical examples, such as a logistic equation with forcing, Duffing oscillator, FitzHugh–Nagumo model, Lorenz attractor and a Susceptible–Infectious–Recovered modeling of SARS-CoV-2.

Combination of ELM and BEMD for Target Recognition of SAR Images

For the problem of target recognition of synthetic aperture radar (SAR) images, a method based on the combination of bidimensional empirical mode decomposition (BEMD) and extreme learning machine (ELM) is proposed. BEMD performs feature extraction for SAR images, producing multi-layer bidimensional intrinsic mode functions (BIMF). These BIMFs covey the discrimination of the original target while effectively eliminating the noises. ELM conducts the classification of each BIMF with high efficiency and robustness. Finally, the decisions from different BIMFs are fused using a linear weighting strategy to reach a reliable decision on the target label. The proposed method compensates the relatively low adaptivity of ELM to noise corruption by BEMD feature extraction. Moreover, the multi-layer BIMFs provide more discriminative information for correct decision. Hence, the overall recognition performance can be improved. As an efficient recognition algorithm, the proposed method can be used in an embedded system for wide applications. Experiments are designed and implemented on the moving and stationary target acquisition and recognition (MSTAR) dataset. The proposed method is tested under both the standard operating condition (SOC) and extended operating conditions (EOCs). The results reflect its effectiveness and robustness via quantitative comparisons.

Learning discriminative representation for image classification

We introduce a new classifier for small-sample image data based on a two-dimensional discriminative regression approach. For a test example, our method estimates a discriminative representation from training examples, which accounts for discriminativeness between classes and enables accurate derivation of categorical information. Unlike existing methods that vectored image data, the learning of the representation in our method is performed with the two-dimensional features of the data, and thus inherent spatial information of the data is fully exploited. This new type of two-dimensional discriminative regression, different from existing regression models, allows for building a highly effective and robust classifier for image data through explicitly incorporating discriminative information and inherent spatial information. We compare our method with several state-of-the-art classifiers of small-sample images and experimental results show superior performance of the proposed method in classification accuracy as well as robustness to noise corruption.

Robustness of transfer learning to image degradation

With the availability of several openly shared large-scale datasets containing millions of images classified in thousands of categories, tremendous progress has been made in image classification research using neural networks in recent years. In real-life image processing applications, however, we often run into situations where the quality of images is less perfect and sample data is limited. This paper investigates the robustness of a transfer learning neural network in the presence of image degeneration. In particular, we examine the tolerance level of the proposed network when there is poor resolution, noise corruption, unwanted cropping, or rotation present in the images; we also explore the range of data size within which the transfer learning algorithm can generate a robust performance. Our study has revealed that the proposed transfer learning algorithm is insensitive to size differences when the sample size reaches a certain threshold. The image processing performances can achieve 75% or higher accuracy in all the image degeneration situations including the worst-case scenario where 100% of the images are corrupted with noise at a 0.5 standard deviation. These experimental results have demonstrated the robustness of the transfer learning neural network. Most importantly, these empirical results lead to a bold suggestion to users in the image processing community. The amount of effort in collecting a huge dataset with high image quality may not be necessary, even just for a fairly precise identification result.

Fast and robust two-dimensional inverse Laplace transformation of single-molecule fluorescence lifetime data

Fluorescence spectroscopy at the single-molecule scale has been indispensable for studying conformational dynamics and rare states of biological macromolecules. Single-molecule two-dimensional (2D) fluorescence lifetime correlation spectroscopy is an emerging technique that holds promise for the study of protein and nucleic acid dynamics, as the technique is 1) capable of resolving conformational dynamics using a single chromophore, 2) resolves forward and reverse transitions independently, and 3) has a dynamic window ranging from microseconds to seconds. However, the calculation of a 2D fluorescence relaxation spectrum requires an inverse Laplace transform (ILT), which is an ill-conditioned inversion that must be estimated numerically through a regularized minimization. Current methods for performing ILTs of fluorescence relaxation can be computationally inefficient, sensitive to noise corruption, and difficult to implement. Here, we adopt an approach developed for NMR spectroscopy (T1-T2 relaxometry) to perform one-dimensional (1D) and 2D-ILTs on single-molecule fluorescence spectroscopy data using singular-valued decomposition and Tikhonov regularization. This approach provides fast, robust, and easy to implement Laplace inversions of single-molecule fluorescence data. We compare this approach to the widely used maximal entropy method.

Parametric Decomposition of Pulsed Lidar Signals with Noise Corruption Using FRFT Spectrum Analysis

The parametric decomposition of full-waveform Lidar data is challenging when faced with heavy noise scenarios. In this paper, we report a fractional Fourier transform (FRFT)-based approach for accurate parametric decomposition of pulsed Lidar signals with noise corruption. In comparison with other joint time-frequency analysis (JTFA) techniques, FRFT is found to present a one-dimensional Lidar signal by a particular two-dimensional spectrum, which can exhibit the mathematical distribution of the multiple components in Lidar signals even with a heavy noise interference. A FRFT spectrum-processing solution with histogram clustering and moving LSM fitting is designed to extract the amplitude, time offset, and pulse width contained in the mathematical distribution. Extensive experimental results demonstrate that the proposed FRFT spectrum analysis method can remarkably outperform the conventional Levenberg–Marquardt-based method. In particular, it can accurately decompose the amplitudes, time offsets, and pulse widths of the pulsed Lidar signal with a −10-dB signal-to-noise-ratio by mean deviation ratios of 4.885%, 0.531%, and 7.802%, respectively.

A Lightweight Fully Convolutional Neural Network for SAR Automatic Target Recognition

Automatic target recognition (ATR) in synthetic aperture radar (SAR) images has been widely used in civilian and military fields. Traditional model-based methods and template matching methods do not work well under extended operating conditions (EOCs), such as depression angle variant, configuration variant, and noise corruption. To improve the recognition performance, methods based on convolutional neural networks (CNN) have been introduced to solve such problems and have shown outstanding performance. However, most of these methods rely on continuously increasing the width and depth of networks. This adds a large number of parameters and computational overhead, which is not conducive to deployment on edge devices. To solve these problems, a novel lightweight fully convolutional neural network based on Channel-Attention mechanism, Channel-Shuffle mechanism, and Inverted-Residual block, namely the ASIR-Net, is proposed in this paper. Specifically, we deploy Inverted-Residual blocks to extract features in high-dimensional space with fewer parameters and design a Channel-Attention mechanism to distribute different weights to different channels. Then, in order to increase the exchange of information between channels, we introduce the Channel-Shuffle mechanism into the Inverted-Residual block. Finally, to alleviate the matter of the scarcity of SAR images and strengthen the generalization performance of the network, four approaches of data augmentation are proposed. The effect and generalization performance of the proposed ASIR-Net have been proved by a lot of experiments under both SOC and EOCs on the MSTAR dataset. The experimental results indicate that ASIR-Net achieves higher recognition accuracy rates under both SOC and EOCs, which is better than the existing excellent ATR methods.

Information capacity and robustness of encoding in the medial prefrontal cortex are modulated by the bioavailability of serotonin and the time elapsed from the cue during a reward-driven task

Serotonin (5-HT) is a key neuromodulator of medial prefrontal cortex (mPFC) functions. Pharmacological manipulation of systemic 5-HT bioavailability alters the electrical activity of mPFC neurons. However, 5-HT modulation at the population level is not well characterized. In the present study, we made single neuron extracellular recordings in the mPFC of rats performing an operant conditioning task, and analyzed the effect of systemic administration of fluoxetine (a selective serotonin reuptake inhibitor) on the information encoded in the firing activity of the neural population. Chronic (longer than 15 days), but not acute (less than 15 days), fluoxetine administration reduced the firing rate of mPFC neurons. Moreover, fluoxetine treatment enhanced pairwise entropy but diminished noise correlation and redundancy in the information encoded, thus showing how mPFC differentially encodes information as a function of 5-HT bioavailability. Information about the occurrence of the reward-predictive stimulus was maximized during reward consumption, around 3 to 4 s after the presentation of the cue, and it was higher under chronic fluoxetine treatment. However, the encoded information was less robust to noise corruption when compared to control conditions.

Towards robust and understandable fault detection and diagnosis using denoising sparse autoencoder and smooth integrated gradients

Industrial applications of fault detection and diagnosis face great challenges as they require not only accurate identification of faulty statuses but also the effective understandability of the results. In this paper, a two-step robust and understandable fault detection and diagnosis framework is developed to address this challenge by exploiting denoising sparse autoencoder and smooth integrated gradients. Specifically, denoising sparse autoencoder(DSAE) is utilized to detect faults in the first step. DSAE is more robust to noise corruption and has better generalization performance compared to the existing autoencoder-based methods. In the second step, smooth integrated gradients(SIG) is used to diagnose the root-cause variables of the faults detected. Smooth integrated gradients can provide a denoising effect on the feature importance. The proposed framework is evaluated through an application to the Tennessee Eastman process. As proved in the experiments, the presented DSAE-SIG method not only achieves higher diagnosis accuracy but also successfully identifies the potential root-cause variables of process disturbances.

Target recognition of synthetic aperture radar images using multi-criteria SRC

ABSTRACT A multi-criteria sparse representation-based classification (SRC) is developed for synthetic aperture radar (SAR) target recognition. The sparse representation is first performed on the testing sample to obtain the coefficient vector over the global dictionary. Accordingly, three different decision criteria are employed for classification, i.e., the minimum global reconstruction error, the minimum local reconstruction error, and the maximum coefficient energy. The three rules exploit the coefficient vector from different aspects so they can complement each other. A linear fusion is conducted on the three result with an adaptive weighting algorithm and a final decision is reached. Based on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset, the proposed method is investigated under the standard operating condition (SOC), noise corruption, and occlusion. The results confirm its validity and robustness.

Data-Driven Fault Detection and Classification for MTDC Systems by Integrating HCTSA and Softmax Regression

The requirement of fast fault isolation poses a great challenge to the safe operation of multi-terminal direct current (MTDC) systems. In order to make a better tradeoff between the speed and reliability of the protection scheme, it is imperative to mine more valuable information from fault transient signals. This paper puts forward a data-driven framework capable of digging out and synthesizing multi-dimensional features to achieve fast and reliable DC fault detection and classification in MTDC systems. Highly comparative time-series analysis (HCTSA) is first adopted to extract extensive features with clear physical interpretations from fault current waveforms, and a few features valuable to fault identification are then selected utilizing the greedy forward search. Based on the reduced features, a softmax regression classifier (SRC) is further proposed to calculate the probability of each fault category with a relatively minor on-line computational burden. Numerical simulations carried out in PSCAD/EMTDC have demonstrated the proposed approach is effective under different fault conditions, robust against noise corruptions as well as abnormal samplings, and replicable in various DC grids. In addition, comprehensive comparison studies with conventional derivative-based protection methods and some typical artificial intelligence based (AI-based) methods have been conducted. It is verified that the proposed method has the advantages of higher fault identification accuracy over conventional protections and shallow structure AI-based methods, better interpretability as well as lower on-line computing complexity over the deep architecture AI-based approaches.

An ERP index of real-time error correction within a noisy-channel framework of human communication

Recent evidence suggests that language processing is well-adapted to noise in the input (e.g., spelling or speech errors, misreading or mishearing) and that comprehenders readily correct the input via rational inference over possible intended sentences given probable noise corruptions. In the current study, we probed the processing of noisy linguistic input, asking whether well-studied ERP components may serve as useful indices of this inferential process. In particular, we examined sentences where semantic violations could be attributed to noise—for example, in “The storyteller could turn any incident into an amusing antidote”, where the implausible word “antidote” is orthographically and phonologically close to the intended “anecdote”. We found that the processing of such sentences—where the probability that the message was corrupted by noise exceeds the probability that it was produced intentionally and perceived accurately—was associated with a reduced (less negative) N400 effect and an increased P600 effect, compared to semantic violations which are unlikely to be attributed to noise (“The storyteller could turn any incident into an amusing hearse”). Further, the magnitudes of these ERP effects were correlated with the probability that the comprehender retrieved a plausible alternative. This work thus adds to the growing body of literature that suggests that many aspects of language processing are optimized for dealing with noise in the input, and opens the door to electrophysiologic investigations of the computations that support the processing of imperfect input.

Synthetic aperture radar target recognition based on joint classification of selected monogenic components by nonlinear correlation information entropy

A synthetic aperture radar (SAR) target recognition method is proposed using monogenic components as basic features. The monogenic signal is employed to decompose original SAR images into multi-scale components. Considering the redundancy and possible indiscrimination in the monogenic components, the nonlinear correlation information entropy (NCIE) is adopted as the criteria for the selection of valid components. The subset of monogenic components with the highest NCIE is chosen and classified by joint sparse representation (JSR). Using the inner correlations of the selected components, JSR could improve the overall reconstruction precision thus enhancing the recognition performance. Experiments are proceeded on the moving and stationary target acquisition and recognition dataset under the standard operating condition and several extended operating conditions, including configuration variances, depression angle variances, noise corruption, and partial occlusion. The results validate the superior effectiveness and robustness of the proposed method over several existed SAR target recognition methods.

Evaluating the impact of acoustic impedance matching on the airborne noise rejection and sensitivity of an electrostatic transducer

To capture sound from solid mediums, such as the human body or musical instruments, transducers designed for airborne sound pickup are typically used. Acoustic impedance mismatches between the medium and transducer decrease the energy captured and increase noise corruption. Here, we demonstrate the improved sensitivity and airborne noise rejection of an electrostatic transducer with an acoustic impedance matched to the medium of interest. The transducer produces an electrical response when mechanical vibrations compress the distance between an elastomer with patterned microstructures and a charged electret film. Using a statistical model generated through an I-optimal design of experiments, the elastomer is fabricated with a specific polymer and concentration of nanoparticles to possess a targeted acoustic impedance. Transducers containing elastomers with impedances in the range of 1 to 2.2 MRayls are assembled and characterized on simulators emulating the human body and a wooden instrument. The noise rejection is quantified using the coherence between the captured transducer signal and the simulated ambient noise. Sensitivities are similarly characterized by comparing the transducer signal and input simulator signal. With an impedance closely matched to the medium of interest, the transducer demonstrates a 2V/N sensitivity and 35 dB SNR improvement compared to an airborne transducer.

Extended convolutional capsule network with application on SAR automatic target recognition

Deep learning-based algorithms have a wide range of applications in synthetic aperture radar (SAR) automatic target recognition (ATR). However, most deep learning-based ATR algorithms perform poorly under extended operation conditions (EOCs), such as large depression variation, noise corruption, limited training samples, partial occlusion, and etc. This paper presents a novel deep neural network named extended convolutional capsule network to achieve robust SAR target recognition under EOCs. The proposed method is composed of an encoder network and a decoder network. Specifically, to mitigate the impact of noise on recognition, we first adopt multiple dilated convolutions to extract multi-scale features in the encoder network. Secondly, a feature refinement module is embedded in the multi-scale channels, which aims to extract discriminative and robust features by adaptively highlighting informative features and suppressing useless ones. Finally, we deploy capsule unit-based feature pose preserving layers to retain the spatial relationship among different features in the top-level network layer of the encoder network, which is very useful for improving the recognition performance under limited training samples. The decoder network is formed by stacking transposed convolutional layers to encourage the encoder network to learn discriminative image features. Experiments on the moving and stationary target acquisition and recognition (MSTAR) data demonstrate the effectiveness of the proposed method.

Design and Analysis of a Signal Denoising and Estimation System

Heart disease is one of the most common causes of death worldwide and electrocardiogram (ECG) signals are among the best tools currently available for diagonising heart disease by offering accurate data about heart performance and the circulatory system. Unfortunately, the linear-quadratic problem often arises, which is a serious problem affecting the estimation of any linear dynamic system’s immediate state due to disruption with white noise due to the linear employment of information about state and white noise corruption. This can theoritecally be addressed by using a Kalman filter as a recursive estimator. However, all biomedical applications that generate electrical signals are disposed to creating different types of noise with fixed coefficients filters, due to the random nature of these signals, and these can cause inaccuracies over time. In this paper, a double Kalman filter is suggested for ECG signal denoising and estimation. Additive White Noise added to the original ECG signal to form a noisy signal, and Kalman filter was used initially to improve the quality of the ECG signal and to offer high performance convergence while increasing the SNR. The second step involved employing Kalman filter to estimate the new denoised ECG signal during a given time period, as accuracy in time is very important in medical applications. The results were satisfactory, offering highly recognisable estimations.