Specific Noise Research Articles

Hardness results for decoding the surface code with Pauli noise

Real quantum computers will be subject to complicated, qubit-dependent noise, instead of simple noise such as depolarizing noise with the same strength for all qubits. We can do quantum error correction more effectively if our decoding algorithms take into account this prior information about the specific noise present. This motivates us to consider the complexity of surface code decoding where the input to the decoding problem is not only the syndrome-measurement results, but also a noise model in the form of probabilities of single-qubit Pauli errors for every qubit.In this setting, we show that quantum maximum likelihood decoding (QMLD) and degenerate quantum maximum likelihood decoding (DQMLD) for the surface code are NP-hard and #P-hard, respectively. We reduce directly from SAT for QMLD, and from #SAT for DQMLD, by showing how to transform a boolean formula into a qubit-dependent Pauli noise model and set of syndromes that encode the satisfiability properties of the formula. We also give hardness of approximation results for QMLD and DQMLD. These are worst-case hardness results that do not contradict the empirical fact that many efficient surface code decoders are correct in the average case (i.e., for most sets of syndromes and for most reasonable noise models). These hardness results are nicely analogous with the known hardness results for QMLD and DQMLD for arbitrary stabilizer codes with independent X and Z noise.

Low-dose radiographic inspection of welding by a novel aperiodic reverse stochastic resonance method

Abstract Low-dose radiographic inspection is a growing trend in industry to minimize radiation risks to humans and the environment. However, reduction in radiation dose often introduces significant noise, which affects image quality and hinders accurate identification of subtle defects. This study addresses this issue by introducing a novel phenomenon called aperiodic reverse stochastic resonance (ARSR), observed in nonlinear systems excited by aperiodic binary signals. ARSR enables simultaneous amplitude amplification and reversal of signals under specific noise conditions. Leveraging ARSR, we propose an image denoising framework for low-dose radiographic inspections. First, a set of projection data is obtained by using Radon transform to reduce the dimensionality of X-ray images from different angles. Then, the projection data is modulated based on the ARSR system. Finally, the image is reconstructed based on the inverse Radon transform. Simulations and experimental comparison results in welding applications validate the effectiveness of the framework, demonstrating significant improvements in image quality for low-dose radiographic defect detection. Unlike advanced methods such as Gaussian filtering, BM3D, and DnCNN, which operate at the pixel level, ARSR performs denoising at the projection data stage, reducing noise impact, preserving original information, and focusing on physical data processing during imaging. This approach enhances the detection of subtle defects, highlighting the potential of stochastic resonance in image processing.

An application of aircraft noise metrics to the Artemis-I launch

In aerospace and acoustical research, there has been significant focus on understanding the negative effects of noise from jet aircraft and flyover vehicles. However, there has been relatively little investigation into the specific noise impacts of launch vehicles. Despite a considerable rise in the number of launches from various spaceports globally in the past decade, there appears to be poor understanding of the potential harm these increasing launches could pose to nearby communities, including both humans and wildlife. This paper aims to apply established noise metrics such as overall sound pressure level, sound exposure level, and effective perceived noise level, commonly used in aircraft policy, to quantify the potential noise impact of launch vehicles, using measurements from the Artemis-I mission as a case study.

How artificial intelligence applied to sound recognition can help gas leak detection in acoustically complex environments?

From production to storage and transport, detecting leaks of flammable gases is a major challenge, particularly in terms of occupational safety and health, and environmental considerations. The emergence of hydrogen as an energy source makes acoustic-based technologies a key element in this value chain. For these reasons, Metravib Engineering, in collaboration with the R&D Safety Project of TotalEnergies One Tech, has been developing since 2019 the AGLED system, for the early detection, localization and classification of these leaks. The system performs acoustic detection in the audible frequency range, using antennas consisting of 4 microphones and physics-based artificial intelligence. Over 5,600 signals were recorded on a site fairly representative of gas production and storage facilities in order to build a robust learning database. Several algorithms were implemented to limit the number of false detections delivered by the system. A relearning strategy was developed to handle unexpected disturbances relative to specific noise environment. This paper describes the overall architecture of the system, from signal acquisition to gas leak detection and 3D localization. It delves into the relearning algorithm and its performance during two full-scale industrial pilots of TotalEnergies - La Mède (France) and Tempa Rossa (Italy) - that included nitrogen controlled leaks tests campaigns.

Classification of BPG-Based Lossy Compressed Noisy Images

Acquired remote sensing images can be noisy. This fact has to be taken into account in their lossy compression and classification. In particular, a specific noise filtering effect is usually observed due to lossy compression and this can be positive for classification. Classification can be also influenced by methodology of classifier learning. In this paper, we consider peculiarities of lossy compression of three-channel noisy images by better portable graphics (BPG) encoder and their further classification. It is demonstrated that improvement of data classification accuracy is not observed if a given image is compressed in the neighborhood of optimal operation point (OOP) and the classifier training is performed for the noisy image. Performance of neural network based classifier is studied. As demonstrated, its training for compressed remote sensing data is able to provide certain benefits compared to training for noisy (uncompressed) data. Examples for Sentinel data used in simulations are offered.

No-reference stereoscopic image quality assessment based on binocular collaboration

Stereoscopic images typically consist of left and right views along with depth information. Assessing the quality of stereoscopic/3D images (SIQA) is often more complex than that of 2D images due to scene disparities between the left and right views and the intricate process of fusion in binocular vision. To address the problem of quality prediction bias of multi-distortion images, we investigated the visual physiology and the processing of visual information by the primary visual cortex of the human brain and proposed a no-reference stereoscopic image quality evaluation method. The method mainly includes an innovative end-to-end NR-SIQA neural network with a picture patch generation algorithm. The algorithm generates a saliency map by fusing the left and right views and then guides the image cropping in the database based on the saliency map. The proposed models are validated and compared based on publicly available databases. The results show that the model and algorithm together outperform the state-of-the-art NR-SIQA metric in the LIVE 3D database and the WIVC 3D database, and have excellent results in the specific noise metric. The model generalization experiments demonstrate a certain degree of generality of our proposed model.

Practical Trainable Temporal Postprocessor for Multistate Quantum Measurement

We develop and demonstrate a trainable temporal postprocessor (TPP) harnessing a simple but versatile machine learning algorithm to provide optimal processing of quantum measurement data subject to arbitrary noise processes for the readout of an arbitrary number of quantum states. We demonstrate the TPP on the essential task of qubit state readout, which has historically relied on temporal processing via matched filters in spite of their applicability for only specific noise conditions. Our results show that the TPP can reliably outperform standard filtering approaches under complex readout conditions, such as high-power readout. Using simulations of quantum measurement noise sources, we show that this advantage relies on the TPP’s ability to learn optimal linear filters that account for general quantum noise correlations in data, such as those due to quantum jumps, or correlated noise added by a phase-preserving quantum amplifier. Furthermore, we derive an exact analytic form for the optimal TPP weights: this positions the TPP as a linearly scaling generalization of matched filtering, valid for an arbitrary number of states under the most general readout noise conditions, all while preserving a training complexity that is essentially negligible in comparison with that of training neural networks for processing temporal quantum measurement data. The TPP can be autonomously and reliably trained on measurement data and requires only linear operations, making it ideal for field-programmable gate array implementations in circuit QED for real-time processing of measurement data from general quantum systems. Published by the American Physical Society 2024

Did Short‐Term Preseismic Crustal Deformation Precede the 2011 Great Tohoku‐Oki Earthquake? An Examination of Stacked Tilt Records

AbstractThe detection of preslip, occurring hours to days before a large earthquake, using geodetic measurements has been a major focus in earthquake prediction research. A recent study claims to have detected a preseismic signal interpreted as accelerating slip near the hypocenter of the 2011 great Tohoku‐oki earthquake, starting approximately 2 hr before the mainshock. This claim is based on a stacking procedure using GNSS (Global Navigation Satellite System) data. However, a follow‐up study demonstrated that the signal disappeared when specific GNSS noise was corrected. Here we utilize tiltmeter records, independent on GNSS, to check whether the claimed preseismic signal is detected using a similar stacking procedure. Our results show no acceleration‐like deformation from 2 hr before the mainshock. This indicates that no precursory slip exceeded the noise level of the tilt data, and if any preslip occurred, it was less than 5.0 × 1018 Nm in seismic moment.

Adaptive noise-aware denoising network: Effective denoising for CT images with varying noise intensity

X-ray computed tomography (CT) plays a crucial role in modern medical imaging for its non-invasive acquisition of anatomical information about the human body. However, its inherent carcinogenic risk remains an unavoidable topic. Low-dose CT (LDCT) reduces ionizing radiation exposure to minimize harm to the human body, but it introduces noise and artifacts of varying intensity, which can affect diagnosis and analysis. In recent years, deep learning has demonstrated competitive performance in medical image denoising. However, most current deep learning-based methods primarily focus on specific noise intensities, leading to degradation when faced with noise of different intensities. In this paper, we propose an Adaptive Noise-Aware Denoising Generative Adversarial Networks (ANAD-GAN) to address the aforementioned issues. The proposed noise-aware memory module can remember and decode high-level hidden features of noise of different intensities adaptively through self-updating feature vectors. To explore a more accurate perception of noise distribution, an intricately designed block named BLOGS is incorporated in our framework. We also integrate a discriminator-based detail-preserving loss in our framework to further enhance the visual quality of the denoised images. In summary, our innovative network dynamically adjusts to varying noise intensities, yielding remarkable denoising results. It attains a PSNR of 48.36 and SSIM of 0.9848 across 5936 images in the AAPM dataset, achieves a PSNR of 48.98 and SSIM of 0.9882 on the Philips dataset with 5000 images, and demonstrates outstanding performance with a PSNR of 50.74 and SSIM of 0.9906 on the private Siemens dataset comprising 4000 images. Through extensive experiments on three datasets, our method surpasses various widely adopted techniques in effectively addressing noise of different intensities.

A novel strategy for detecting mutations from circulating tumor DNA with blood tissue only.

e20031 Background: Molecular Residual Disease (MRD) test detects tumor-specific mutations in circulating tumor DNA (ctDNA), which can indicate the risk of tumor recurrence. However, without tumor-specific mutations, detecting mutations in blood tissue alone is extremely challenging due to the low concentration of ctDNA in blood. To address this challenge, we proposed an innovative mutation filtering and calling strategy by simulating noise’s distribution and quality assessment to greatly boost the performance. Methods: We performed 406 pairs of MRD tests on both tumor and blood tissue of postoperative patients with mixed cancer types, including lung, gastrointestinal (GI), hepatobiliary, and pancreatic. Patients that possess one or more tumor-specific mutations in their blood tissue were classified as recurrence high-risk (n = 66), others were classified as recurrence low-risk (n = 340). All sequence data were analyzed using both traditional and proposed method. In the traditional method, VAF threshold, end-reads, multiple-align reads and population frequencies were used as filtering criteria. While in the proposed method, VAF threshold was excluded and two new parameters were introduced: The distribution of variant allele frequency (VAF) of sequencing noise, which simulated 42 healthy individuals’ background noise with conf > 0.995; The normalized proportion of supported unique and total variants, to assess the likelihood of mutations originated from ctDNA. Results: Among 406 patients of mixed tumor types, the traditional approach barely reached a sensitivity of 39.4% and a specificity of 89.1%, while our strategy achieved a sensitivity of 71.2% and a specificity of 80.9%. Particularly among the 92 patients with gastrointestinal tumor, by filtering specific noise caused by the MRD panel (e.g. multi-alignment error of MUC gene family), we further boost the sensitivity from 47.1% to 82.4% with a loss of specificity from 93.3% to 90.7%. Moreover, among 248 patients with lung tumors, the sensitivity and specificity between traditional approach and proposed method were 15.8%, 93.9%, and 52.6%, 90.0% respectively, which might be caused by the low recurrence rate in the patient cohort (19 out of 248). Conclusions: The novel strategy based on simulated noise distribution and read quality assessment surpasses the traditional threshold-based approaches in mutation detection, especially in deep sequencing like MRD tests. The findings of the study indicate that the recurrence monitoring strategy is applicable across various tumor types and exhibits superior performance in GI tumor, albeit with certain limitations. In conclusion, this study boosts the feasibility of MRD as a non-invasive alternative for tissue biopsy by providing a novel mutation filtering strategy, which further strengthen the potential of MRD testing in enhancing cancer patients’ long-term prognosis and treatment.

Analyzing VNO Airport Traffic Data of 2023: Specific Aircraft Noise Measurement and Mitigation Recommendations

The expansion of airport operations worldwide yields numerous advantages, yet it also brings about certain adverse consequences, notably noise pollution. In addressing noise pollution, airports deploy a range of noise control strategies, which can be broadly classified into four categories: passive noise mitigation methods, active noise mitigation strategies, regulatory measures, and technological innovations. The paper investigates the feasibility of Balanced Approach (BA) implementation at Vilnius International Airport (VNO according to IATA). In this study, an in depth analysis was conducted on the data pertaining to the frequency of movements, the distinct types, models and age of aircraft that utilized VNO facilities during the calendar year of 2023. Following the analysis conducted throughout 2023, a total of 38 699 aircraft movements were recorded, from which predominant aircraft types were identified: Boeing B737 (A1), Airbus A320 (A2), Airbus A220 (A3), Bombardier CRJ/Challenger (A4), Embraer (A5) and ATR 72-500 (A6). Average age of aircraft most frequently arriving and departing at VNO is 14.35 years. The sound measurements for A1, A3 and A6 were done with Bruel&Kjaer 2270 Investigator sound analyzer. Data then was post processed by BZ-5503 Measurement Partner Suite and measured data Leq, LMAX, different noise levels were compared. Measurements show that permitted noise levels are exceeded, especially by older age aircraft. Reducing air fleet age would be most efficient noise mitigation measure according to BA.

A Contribution-Aware Noise Feature representation model for image manipulation localization

Noise feature modules play a prevalent role in image manipulation localization. However, existing approach involves utilizing a specific noise feature module tailored to address distinct categories of tampering procedures, limiting its adaptability across diverse scenarios. This limitation becomes particularly pronounced in image manipulation localization where a singular noise module struggles to comprehensively represent all potential tampering types, given the inherent uncertainty surrounding the nature of tampering. A novel Contribution-Aware Noise Feature (CANF) representation model, specifically designed to tackle the intricacies of image tamper noise, is introduced in this paper. The proposed model integrates existing noise feature modules, autonomously enhancing their capabilities. This enhancement not only contributes significantly to image manipulation localization during the inference phase but also enables a quantitative evaluation of the distinct contributions of various noise modules to the inference results. To optimize the utilization of image information, the Red–Green–Blue (RGB) and CANF data are fused, leading to the development of a standardized image manipulation localization architecture. This architecture exhibits heightened efficiency in harnessing diverse types of information, thereby enhancing the overall performance of image manipulation localization. Experimental results demonstrate that our approach, alongside established methods IF-OSN, MSFF, PCL, and NCL, surpasses the state-of-the-art NCL by 8.5% and 5.4% concerning average F1 and AUC metrics across four standard image manipulation localization datasets: NIST16, CASIAv1, Columbia, and Coverage. In addition, we comprehensively explore the impact of various noise characterization modules on different types of tampering and the fact that Late Fusion is more conducive to image tampering localization.

Reducing construction noise: sound masking effect on soundscape dominated by construction noise

AbstractPeople move to cities to enjoy the conveniences of modern life, but the resulting population growth comes with issues such as increased noise. While public awareness of noise pollution is increasing, noise emanating from construction sites remains a source of complaint. Despite the actions taken by different parties to control construction noise, newly constructed structures continue to exacerbate the harmful consequences of noise. The acoustic environment in some areas of cities, such as London, can no longer provide the tranquility expected by residents, and the risk of daily disruptions for those living near construction sites is high. Given the various limitations and complications of current construction noise control methods, the campaign to limit construction noise must embrace other strategies. Sound masking can potentially abate noise without resorting to time-consuming procedures. In this paper, five raw construction noise datasets were collected around an existing noise-sensitive premise. Five masker sound datasets were recorded to form seven mixed sounds to create a comparative experiment on masking construction noise. A case study and additional research are presented to demonstrate the mechanism of sound masking and to investigate the factors involved in effectively masking construction noise using the sound masking concept. Based on the case study and theoretical research findings, suggestions for masker selection of specific noise datasets and conclusions regarding ideal sound simultaneous masking of construction noise are provided.

Oversampling framework based on sample subspace optimization with accelerated binary particle swarm optimization for imbalanced classification

In response to the need to generate synthetic minority class samples to extend minority classes, the SMOTE-based oversampling methods have been favored for class-imbalanced classification. They usually generate unnecessary noise when training data are not well separated. Although filtering-based oversampling methods are recognized as effective solutions for addressing noise generation through employing specific noise filters based on instance selection methods to remove suspicious noise, they suffer from the following issues: a) noise filters heavily rely on strong assumptions, causing low robustness to different datasets; b) noise filters are specially designed for a specific oversampling method, and are not easily extended to others; and c) noise filters have a relatively high time consumption. To address noise generation while overcoming the above issues a)-c), an oversampling framework based on sample subspace optimization with accelerated binary particle swarm optimization (OF-SSO-ABPSO) is proposed. OF-SSO-ABPSO is a wrapping framework compatible with almost all the oversampling methods. First, in the framework, a SMOTE-based method is used to generate synthetic minority class samples. Second, a novel accelerated binary particle swarm optimization (ABPSO) algorithm with a new search space reduction strategy, a new particle update mechanism, and a new fitness function is proposed. Third, a novel ABPSO-based sample subspace optimization (SSO-ABPSO) method is proposed and used as a noise filter to remove suspicious noise from the training set and synthetic minority class samples. Experiments prove that, a) OF-SSO-ABPSO can improve 6 representative SMOTE variations by addressing noise generation, and b) OF-SSO-ABPSO outperforms 7 state-of-the-art filtering-based oversampling methods.

Combined Cubature Kalman and Smooth Variable Structure Filtering Based on Multi-Kernel Maximum Correntropy Criterion for the Fully Submerged Hydrofoil Craft

This paper introduces a novel filter algorithm termed as an MKMC-CSVSF which combined square-root cubature Kalman (SR-CKF) and smooth variable structure filtering (SVSF) under multi-kernel maximum correntropy criterion (MKMC) for accurately estimating the state of the fully submerged hydrofoil craft (FSHC) under the influence of uncertainties and multivariate heavy-tailed non-Gaussian noises. By leveraging the precision of the SR-CKF and the robustness of the SVSF against system uncertainties, the MKMC-CSVSF integrates these two methods by introducing a time-varying smooth boundary layer along with multiple fading factors. Furthermore, the MKMC is introduced for the adjustment of kernel bandwidths across different channels to align with the specific noise characteristics of each channel. A fuzzy rule is devised to identify the appropriate kernel bandwidths to ensure filter accuracy without undue complexity. The precision and robustness of state estimation in the face of heavy-tailed non-Gaussian noises are improved by modifying the SR-CKF and the SVSF using a fixed-point approach based on the MKMC. The experimental results validate the efficacy of this algorithm.

Bidirectional Multi-Spectral Vibration Control: Insights from Automotive Engine Mounting Systems in Two-Dimensional Structures with a Damaged Vertical Active Element

Active mounting systems have become more prevalent in recent years to effectively mitigate structure-induced vibration across the automobile chassis. This trend is particularly evident in engine mounts. Considerable research has been dedicated to this approach owing to its potential to enhance the quietness and travel comfort of automobiles. However, prior research has concentrated on a limited spectrum of specific vibrations and noise control or has been restricted to vertical vibration control. This article describes the modeling, analysis, and control of a source structure employing a multidirectional active mounting system designed to closely simulate the position and direction of an actual automobile engine mount. A piezoelectric stack actuator is connected in series to an elastic (rubber) mount to form an active mount. The calculation of the secondary force required for each active mount is achieved through the application of harmonic excitation forces. The control signal can also reduce vibrations caused by destructive interference with the input signal. Furthermore, horizontal oscillations can be mitigated by manipulating the parameters via dynamic interconnections of the source structure. We specifically examined the level of vibration reduction performance in the absence of a vertical active element operation and determined whether the control is feasible. Simulation outcomes demonstrate that this active mount, which operates in both the vertical and horizontal directions, effectively mitigates excitation vibrations. Furthermore, a simulation was conducted to mitigate the vibrations caused by complex signals (AM and FM signals) and noise. This was achieved by monitoring the system response using an adaptive filter NLMS algorithm. Adaptive filter simulations demonstrate that the control efficacy degrades in response to complex signals and noise, although the overall relaxation trend remains unchanged.

Comparative analysis of ammonia and hydrogen fuel blends in diesel engines on performance, emission, vibration, and acoustic profiles

In the current study two kinds of gaseous fuels, ammonia and hydrogen tested along with the neat diesel. A series of test conducted in the diesel engine for the samples such as diesel, S1, S2, S3 and S4. The samples were varied based on the two different flow rates 5 LPM and 10 LPM. The parameters such as brake thermal efficiency (BTE), brake specific energy consumption (BSEC), combustion parameters, emissions, and noise levels were determined. The research suggests that the typical properties of ammonia and hydrogen blends contribute to engine performance and efficiency. The S4 blend, 10 LPM of NH3 and 10 LPM of H2, with a higher hydrogen content, was found to significantly enhance BTE across all engine brake powers compared to conventional diesel fuel. A notable increase in BTE with a climb on the brake power was observed, suggesting superior fuel energy utilization due to the blend's high-energy content. An inverse relationship was established between BTE and BSEC, where diesel demonstrated higher BSEC, indicative of greater thermal loss and inefficiency. Combustion analysis revealed a reduction in heat release rate for ammonia blends, this eventually compensated by hydrogen addition by optimizing autoignition and accelerating combustion. Emission studies highlighted a uniform increase in nitrogen oxides for all blends; however, the combination of hydrogen with ammonia was shown to mitigate NOx formation by lowering flame temperatures. Additionally, vibrational analysis demonstrated that ammonia and hydrogen blends reduce engine vibration, especially at higher loads, and contribute to decreased combustion and exhaust noise.

Efficient Optimization of a Support Vector Regression Model with Natural Logarithm of the Hyperbolic Cosine Loss Function for Broader Noise Distribution

While traditional support vector regression (SVR) models rely on loss functions tailored to specific noise distributions, this research explores an alternative approach: ε-ln SVR, which uses a loss function based on the natural logarithm of the hyperbolic cosine function (lncosh). This function exhibits optimality for a broader family of noise distributions known as power-raised hyperbolic secants (PHSs). We derive the dual formulation of the ε-ln SVR model, which reveals a nonsmooth, nonlinear convex optimization problem. To efficiently overcome these complexities, we propose a novel sequential minimal optimization (SMO)-like algorithm with an innovative working set selection (WSS) procedure. This procedure exploits second-order (SO)-like information by minimizing an upper bound on the second-order Taylor polynomial approximation of consecutive loss function values. Experimental results on benchmark datasets demonstrate the effectiveness of both the ε-ln SVR model with its lncosh loss and the proposed SMO-like algorithm with its computationally efficient WSS procedure. This study provides a promising tool for scenarios with different noise distributions, extending beyond the commonly assumed Gaussian to the broader PHS family.

Influences of parameters of film on acoustical absorption performance of copper film-copper fiber composite structure

Fiber porous materials excel at sound absorption, but their effectiveness suffers at thin thicknesses. This study proposed a novel copper film-copper fiber porous material (FFM) structure to overcome this limitation. We designed a composite material with improved sound absorption, by coating a thin copper film onto an existing copper fiber matrix. This innovative approach enables effective sound absorption when the sample thickness is small. Our analysis revealed that FFM structures are better than the individual copper film and fiber porous material regarding sound absorption, particularly near their resonance frequency. Notably, a 2 mm-thick FFM structure boasts an average sound absorption coefficient of 0.35, a remarkable fourfold increase compared to the individual copper fiber matrix (0.08). The further investigation explores the influence of copper film thickness and bond area percentage on acoustic performance. We successfully validate our simulation model against experimental data, enabling us to calculate energy consumption ratios for various design parameters. This provides research ideas for optimizing FFM structures for specific noise reduction applications.

Cross noise level PET denoising with continuous adversarial domain generalization

Objective. Performing positron emission tomography (PET) denoising within the image space proves effective in reducing the variance in PET images. In recent years, deep learning has demonstrated superior denoising performance, but models trained on a specific noise level typically fail to generalize well on different noise levels, due to inherent distribution shifts between inputs. The distribution shift usually results in bias in the denoised images. Our goal is to tackle such a problem using a domain generalization technique. Approach. We propose to utilize the domain generalization technique with a novel feature space continuous discriminator (CD) for adversarial training, using the fraction of events as a continuous domain label. The core idea is to enforce the extraction of noise-level invariant features. Thus minimizing the distribution divergence of latent feature representation for different continuous noise levels, and making the model general for arbitrary noise levels. We created three sets of 10%, 13%–22% (uniformly randomly selected), or 25% fractions of events from 97 18F-MK6240 tau PET studies of 60 subjects. For each set, we generated 20 noise realizations. Training, validation, and testing were implemented using 1400, 120, and 420 pairs of 3D image volumes from the same or different sets. We used 3D UNet as the baseline and implemented CD to the continuous noise level training data of 13%–22% set. Main results. The proposed CD improves the denoising performance of our model trained in a 13%–22% fraction set for testing in both 10% and 25% fraction sets, measured by bias and standard deviation using full-count images as references. In addition, our CD method can improve the SSIM and PSNR consistently for Alzheimer-related regions and the whole brain. Significance. To our knowledge, this is the first attempt to alleviate the performance degradation in cross-noise level denoising from the perspective of domain generalization. Our study is also a pioneer work of continuous domain generalization to utilize continuously changing source domains.