Denoising Performance Research Articles

Unsupervised Cryo-EM Images Denoising and Clustering based on Deep Convolutional Autoencoder and K-Means+.

Cryo-electron microscopy (cryo-EM) is a widely used structural determination technique. Because of the extremely low signal-to-noise ratio (SNR) of images captured by cryo-EM, clustering single-particle cryo-EM images with high accuracy is challenging. To address this, we proposed an iterative denoising and clustering method based on a deep convolutional variational autoencoder and K-means++. The proposed method contains two modules: a denoising ResNet variational autoencoder (DRVAE), and Balance size K-means++ (BSK-means++). First, the DRVAE is trained in a fully unsupervised manner to initialize the neural network and obtain preliminary denoised images. Second, BSK-means++ is built for clustering denoised images, and images closer to class centers are divided into reliable samples. Third, the training of DRVAE is continued, while the class-average images are used as pseudo supervision of reliable samples to reserve more detailed information of denoised images. Finally, the second and third steps mentioned above can be performed jointly and iteratively until convergence occurs. The experimental results showed that the proposed method can generate reliable class average images and achieve better clustering accuracy and normalized mutual information than current methods. This study confirmed that DRVAE with BSK-means++ could achieve a good denoise performance on single-particle cryo-EM images, which can help researchers obtain information such as symmetry and heterogeneity of the target particles. In addition, the proposed method avoids the extreme imbalance of class size, which improves the reliability of the clustering result.

Optimization of Signal Space Separation for Optically Pumped Magnetometer in Magnetoencephalography.

Magnetoencephalography (MEG) is a noninvasive functional neuroimaging modality but highly susceptible to environmental interference. Signal space separation (SSS) is a method for improving the SNR to separate the MEG signals from external interference. The origin and truncation values of SSS significantly affect the SSS performance. The origin value fluctuates with respect to the helmet array, and determining the truncation values using the traversal method is time-consuming; thus, this method is inappropriate for optically pumped magnetometer (OPM) systems with flexible array designs. Herein, an automatic optimization method for the SSS parameters is proposed. Virtual sources are set inside and outside the brain to simulate the signals of interest and interference, respectively, via forward model, with the sensor array as prior information. The objective function is determined as the error between the signals from simulated sources inside the brain and the SSS reconstructed signals; thus, the optimized parameters are solved inversely by minimizing the objective function. To validate the proposed method, a simulation analysis and MEG auditory-evoked experiments were conducted. For an OPM sensor array, this method can precisely determine the optimized origin and truncation values of the SSS simultaneously, and the auditory-evoked component, for example, N100, can be accurately located in the temporal cortex. The proposed optimization procedure outperforms the traditional method with regard to the computation time and accuracy, simplifying the SSS process in signal preprocessing and enhancing the performance of SSS denoising.

Low-rank with sparsity constraints for image denoising

Recent works on providing proper sparse or low-rank priors have shown to result in good quality image restoration performance. The nonlocal self-similarity (NSS) of images indicates that groups stacked by similar patches not only have low-rank characteristic, but also have sparse characteristic. However, many of the existing methods for image modeling use one or the other of these characteristics, but not both, which may limit the performance of image denoising. We propose in this paper a novel model for image denoising, designated as the bilateral weighted sparse coding and low-rank (BWSCLR) model, which exploits simultaneously the low-rank property and the bilateral weighted sparse coding under the NSS prior to better restore the texture and the edges of images contaminated by noise. Furthermore, a block coordinate descent algorithm involving three steps is proposed to obtain the solution for our model. A number of experiments on denoising different image scenarios, including gray-scale and color images are carried out, and the results clearly show the superior performance of the proposed BWSCLR method over that of the state-of-the-art algorithms not only in terms of the quantitative metrics such as PSNR and SSIM, but also in perceptual quality.

Live 4D-OCT denoising with self-supervised deep learning

By providing three-dimensional visualization of tissues and instruments at high resolution, live volumetric optical coherence tomography (4D-OCT) has the potential to revolutionize ophthalmic surgery. However, the necessary imaging speed is accompanied by increased noise levels. A high data rate and the requirement for minimal latency impose major limitations for real-time noise reduction. In this work, we propose a low complexity neural network for denoising, directly incorporated into the image reconstruction pipeline of a microscope-integrated 4D-OCT prototype with an A-scan rate of 1.2 MHz. For this purpose, we trained a blind-spot network on unpaired OCT images using a self-supervised learning approach. With an optimized U-Net, only a few milliseconds of additional latency were introduced. Simultaneously, these architectural adaptations improved the numerical denoising performance compared to the basic setup, outperforming non-local filtering algorithms. Layers and edges of anatomical structures in B-scans were better preserved than with Gaussian filtering despite comparable processing time. By comparing scenes with and without denoising employed, we show that neural networks can be used to improve visual appearance of volumetric renderings in real time. Enhancing the rendering quality is an important step for the clinical acceptance and translation of 4D-OCT as an intra-surgical guidance tool.

Edge-preserving smoothing filter using fast M-estimation method with an automatic determination algorithm for basic width

Digital filtering is essential for digital imaging, image recognition, and super-resolution technology. For example, the presence of noise in images captured by digital cameras causes deterioration of the image quality and image recognition rate. In order to improve the image recognition rate, noise reduction and edge preservation must be performed during preprocessing. Noise is generally reduced using low-pass filters, such as the Gaussian filter. Although they reduce noise, such filters also have the properties of blurring edge. A strong edge blur reduces the accuracy of the feature detection in image recognition. Therefore, in our previous study, a fast M-estimation Gaussian filter for images (FMGFI) was proposed as an image filter that simultaneously achieves denoising and edge preservation. In the FMGFI, the setting of the optimal basic width of the 2nd order B-spline basis functions is important for achieving simultaneous denoising and edge preservation. In this method, the optimal basic width of the FMGFI was determined not only by manually setting the basic width but also by human judgment of the filtered images. Consequently, the inability to automatically determine the optimal basic width hindered efficient denoising during image processing Therefore, in this research, we develop and propose a method that can automatically determine the optimal basic width of the FMGFI. The previously proposed method calculates using the same basic width for all the pixels over the entire image; in contrast, the proposed method calculates using the basic width automatically determined for each pixel. The experiments confirmed that the method proposed in this study achieves higher denoising and edge preservation performance than the ones used in previous research. The results also showed that it has the highest denoising performance against salt-and-pepper noise as compared to other filters: non-local mean filter, Gaussian filter, median filter, bilateral filter, adaptive bilateral filter, and FMGFI. The experimental results for the Gaussian noise sowed that the proposed method has the same denoising and edge preservation performance as the other filters in visual evaluation. From the above, the proposed method is expected to contribute to efficient denoising and improvement of image quality by using it as a preprocessing.

Low-Dose CT Denoising via Sinogram Inner-Structure Transformer.

Low-Dose Computed Tomography (LDCT) technique, which reduces the radiation harm to human bodies, is now attracting increasing interest in the medical imaging field. As the image quality is degraded by low dose radiation, LDCT exams require specialized reconstruction methods or denoising algorithms. However, most of the recent effective methods overlook the inner-structure of the original projection data (sinogram) which limits their denoising ability. The inner-structure of the sinogram represents special characteristics of the data in the sinogram domain. By maintaining this structure while denoising, the noise can be obviously restrained. Therefore, we propose an LDCT denoising network namely Sinogram Inner-Structure Transformer (SIST) to reduce the noise by utilizing the inner-structure in the sinogram domain. Specifically, we study the CT imaging mechanism and statistical characteristics of sinogram to design the sinogram inner-structure loss including the global and local inner-structure for restoring high-quality CT images. Besides, we propose a sinogram transformer module to better extract sinogram features. The transformer architecture using a self-attention mechanism can exploit interrelations between projections of different view angles, which achieves an outstanding performance in sinogram denoising. Furthermore, in order to improve the performance in the image domain, we propose the image reconstruction module to complementarily denoise both in the sinogram and image domain.

Parallel Processing Method for Microseismic Signal Based on Deep Neural Network

The microseismic signals released by rock mass fracture can be captured via microseismic monitoring to evaluate the development of geological disasters. This is crucial for underground engineering construction, underground mining, and earthquake and geological disaster evaluation. However, extracting information effectively is difficult due to the low signal-to-noise ratio of microseismic signals caused by complex environmental factors. Therefore, denoising and detection (onset time picking) are essential to processing microseismic signals and extracting source information. To improve the efficiency and accuracy of microseismic signal processing, we propose a parallel dual-tasking network, which is an advanced deep learning model that can simultaneously perform microseismic denoising and detection tasks. The network, comprising one encoder and two parallel decoders, is customised to extract input data features, and two outputs can be simultaneously generated to denoise and detect microseismic signals. The model exhibits excellent denoising and detection performance for microseismic signals containing various types of noise. Compared with traditional methods, the signal-to-noise ratio of the denoised signal is greatly improved, and the waveform distortion of the denoised signal is small. Even when the signal-to-noise ratio is low, the proposed model can maintain good onset time pickup performance. This method obviates the need for different denoising methods for different types of noise and precludes setting thresholds artificially to improve the denoising effect and detection accuracy. Moreover, the dual processing characteristics of the model facilitate simultaneous denoising and detection, which improves the processing efficiency of microseismic data and meets the demand for automatically processing massive microseismic data. Therefore, this method has excellent data processing potential in exploration seismology, and earthquake and disaster assessment.

Convolutional Neural Network Denoising Auto-Encoders for Intelligent Aircraft Engine Gas Path Health Signal Noise Filtering

Abstract Removing noise from health signals is critical in gas path diagnostics of aircraft engines. An efficient noise filtering/denoising method should remove noise without using future data points, preserve important changes, and promote accurate diagnostics without time delay. Machine learning (ML)-based methods are promising for high fidelity, accuracy, and computational efficiency under the motivation of Intelligent Engines. However, previous ML-based denoising methods are rarely applied in actual engineering practice because they cannot accommodate time series and cannot effectively capture important changes or are limited by the time delay problem. This paper proposes a convolutional neural network denoising auto-encoder (CNN-DAE) method to build a denoising auto-encoder structure. In this structure, a convolutional operation is used to accommodate time series, and causal convolution is introduced to solve the problem of using future data points. The proposed denoising method is evaluated against NASA's propulsion diagnostic method evaluation strategy (prodimes) software. It has been proved that the proposed method can accommodate time series, remove noise for improved denoising accuracy, and preserve the important changes for enhanced diagnostic information. NASA's blind test case results show that kappa coefficient of a common diagnostic method using the processed data is 0.731 and is at least 0.046 higher than the other diagnostic methods in the open literature. Processing health signals using the proposed method would significantly promote accurate diagnostics without time delay. The proposed method could support intelligent condition monitoring systems by exploiting historical information for improved denoising and diagnostic performance.

Boosting deep image prior by integrating external and internal image priors

Recently, an unsupervised method [denoted as deep image prior (DIP)] that interprets images as the output of a convolutional network with randomly initialized inputs has received much attention, as the DIP can adapt its network parameters to any given noisy image without requiring training pairs, providing a great flexibility for handling those noisy images with complex structures and low degree of self-similarity. However, the DIP method still suffers from a crucial limitation that only the noisy image is used as target image, which implies that only internal prior information is used for noise removal and leads to poor denoising performance, compared with supervised denoising models. In this work, we aim to boost the DIP method by integrating external and internal image priors. Specifically, with a given noisy image, we first exploited a state-of-the-art supervised method to denoise it and obtain its corresponding denoised image called initial denoised image. Then the initial denoised image containing rich external prior information was used as extra target image together with the given noisy image within the standard DIP framework, resulting in a highly effective unsupervised recovery process. Next, to take full advantage of the uncertainty of the DIP network, the above DIP denoising routine with the different random inputs was executed multiple times to generate enough complementary denoised images (samples). Finally, with an unsupervised weight map generative network, the generated samples were fused in a pixel-wise manner. The fused image with better image quality was treated as the final denoised image. We had verified the denoising performance of our method on a large of images from several benchmark databases as well as real-world noisy images. The comparative results show that our method outperforms the original DIP method and other unsupervised methods by a large margin, and surpasses state-of-the-art supervised counterparts with comparable peak signal-to-noise rate. Source code is publicly available at https://github.com/chenxiaojun0101/BDIP.

Research on Image Denoising in Edge Detection Based on Wavelet Transform

Photographing images is used as a common detection tool during the process of bridge maintenance. The edges in an image can provide a lot of valuable information, but the detection and extraction of edge details are often affected by the image noise. This study proposes an algorithm for wavelet transform to denoise the image before edge detection, which can improve the signal-to-noise ratio of the image and retain as much edge information as possible. In this study, four wavelet functions and four decomposition levels are used to decompose the image, filter the coefficients and reconstruct the image. The PSNR and MSE of the denoised images were compared, and the results showed that the sym5 wavelet function with three-level decomposition has the best overall denoising performance, in which the PSNR and MSE of the denoised images were 23.48 dB and 299.49, respectively. In this study, the canny algorithm was used to detect the edges of the images, and the detection results visually demonstrate the difference between before and after denoising. In order to further evaluate the denoising performance, this study also performed edge detection on images processed by both wavelet transform and the current widely used Gaussian filter, and it calculated the Pratt quality factor of the edge detection results, which were 0.53 and 0.47, respectively. This indicates that the use of wavelet transform to remove noise is more beneficial to the improvement of the subsequent edge detection results.

Evaluation of edge detection algorithm of frontal image of facial contour in plastic surgery

With the improvement of medical levels and the continuous improvement of people’s living standards, the demand for beauty by the general public is increasing. The plastic surgery industry has also developed by leaps and bounds. People’s dissatisfaction with their own facial appearance, facial injuries and some other reasons have prompted people to carry out facial reconstruction, and facial plastic surgery has developed rapidly. However, in the current facial plastic surgery, the edge detection effect on the contour image is general. In order to improve the edge detection effect of facial contour lines in medical images, this paper proposed a facial contour line generation algorithm. First, the detection effects of four operators were compared. After comparing the effects, the Sobel operator was used as the input data to generate an edge detection algorithm. Then, the grayscale features of the tissue in the image and the symmetry of the image were used to perform bidirectional contour tracking on the detected image to extract facial contour lines. In addition, for facial contour features, the midpoint method can be used to generate auxiliary contours. The algorithm was verified by a set of facial CT (Computed Tomography) images in the experiment. The results showed that the new generation algorithm accelerated the edge detection speed, had good denoising performance, and enhanced the edge detection effect by about 12.05% compared with the traditional edge detection algorithm. The validity and practicability of facial edge detection were verified, and it provided a theoretical basis for further realizing the design of a facial contour digital image processing system.

A multi-scale dense-connection denoising network for DAS-VSP records

Due to high spatial resolution, low cost, and wide bandwidth, distributed optical fiber acoustic sensing (DAS) is regarded as a potential tool for data acquisition in vertical seismic profile (VSP) surveys. However, in real DAS-VSP records, desired signals are often seriously plagued by various noise, which does not appear in the conventional seismic data received by electronic geophones. Exploring a high-performing attenuation method for the background noise can significantly improve the quality of DAS-VSP records and has essential impacts on the following imaging and interpretation. Deep-learning-based methods, especially convolutional neural network (CNN), have shown remarkable performance in seismic data denoising. However, the conventional CNN-based methods may degrade when dealing with DAS-VSP records in low signal-to-noise ratio. In this study, we propose a novel multi-scale dense-connection denoising network (MDD-Net) to achieve high-accuracy processing of the complex DAS background noise. Unlike conventional multi-scale networks, MDD-Net utilizes widen convolution block to capture the multi-scale features of the analyzed data. On this basis, dense connection operations are employed to fuse the features and improve the network efficiency. Meanwhile, an enhanced spatial attention (ESA) block is designed to reinforce the features, which are helpful for noise suppression and weak signal recovery. Both synthetic and field DAS-VSP records are processed to verify the effectiveness of MDD-Net. Meanwhile, we also compare the denoising results with other competing methods. The experimental results demonstrate that MDD-Net can significantly attenuate the complex DAS background noise and restore the desired signals, even for the weak upgoing signals.

MD3: Model-Driven Deep Remotely Sensed Image Denoising

Remotely sensed images degraded by additive white Gaussian noise (AWGN) have low-level vision, resulting in a poor analysis of their contents. To reduce AWGN, two types of denoising strategies, sparse-coding-model-based and deep-neural-network-based (DNN), are commonly utilized, which have their respective merits and drawbacks. For example, the former pursue enjoyable performance with a high computational burden, while the latter have powerful capacity in completing a specified task efficiently, but this limits their application range. To combine their merits for improving performance efficiently, this paper proposes a model-driven deep denoising (MD3) scheme. To solve the MD3 model, we first decomposed it into several subproblems by the alternating direction method of multipliers (ADMM). Then, the denoising subproblems are replaced by different learnable denoisers, which are plugged into the unfolded MD3 model to efficiently produce a stable solution. Both quantitative and qualitative results validate that the proposed MD3 approach is effective and efficient, while it has a more powerful ability in generating enjoyable denoising performance and preserving rich textures than other advanced methods.

Multi-scale interactive network in the application of DAS seismic data processing

Distributed acoustic sensing (DAS) is regarded as a novel acquisition technology for seismic data. Compared with conventional electrical geophones, DAS has a series of obvious advantages including low-cost, high spatial resolution, good coverage, and strong resistance to the harsh environment. Noise attenuation is an essential step in seismic data processing. However, there are two main difficulties faced by the denoising task of DAS seismic data. On the one hand, some background noise in DAS seismic data, such as optical low-frequency noise, horizontal noise, and fading noise, is unique and not presented in the conventional seismic data; on the other hand, the signal-to-noise ratio (SNR) of DAS seismic data is relatively low. Recently, a convolutional neural network (CNN) has shown superior denoising performance compared to the traditional method. To follow this promising trend, we propose a multi-scale interactive convolutional neural network (MSI-Net) and apply it to denoise the challenging DAS seismic data. Different from most of the existing CNN architecture used in seismic data denoising, the MSI-Net considers both coarse-scale and fine-scale features by improving the inherent serial convolution to multi-scale parallel convolution, which is beneficial to recover detailed information. Moreover, we utilize some connections to achieve the information interaction between different scales, which promotes the flow of information and enables the network to extract more informative multi-scale features from the DAS seismic data. Moreover, both synthetic and real examples demonstrate that the proposed MSI-Net can effectively attenuate a variety of unique DAS background noise and also completely recover the weak signals. Compared with conventional CNN architecture, MSI-Net exhibits better performance in global SNR and local details.

Interpretable Poisson Optimization-Inspired Deep Network for Single-Photon Counting Image Denoising

Single-photon counting (SPC) imaging is a versatile approach for detecting targets under extremely low-light situations. To increase the quality of SPC images degraded by noise, traditional optimization-based methods seek priors from handcrafted features, which are inadequate to handle different kinds of targets in natural scenes, leading to poor denoising performance. The difficulty of optimal parameter design for good balance between denoising performance and spatial clarity is another obstacle. To address these issues, we develop a novel interpretable Poisson optimization-inspired deep network for SPC image denoising. First, we build a learnable prior that regularizes the Poisson optimization problem for SPC imaging to enhance the denoising performance. We construct a deep network by unfolding the iterative shrinkage-thresholding algorithm to solve the Poisson optimization problem. Therefore, all modules in the network have strong interpretability, enabling good generalization capability in real situations. Second, all parameters are optimized in a data-driven manner in the network. Finally, we conduct both simulated and real experiments to test the effectiveness of the proposed method. Experimental results demonstrate that the proposed method outperforms other state-of-the-art methods from aspects of visual effect and quantitative analysis.

Versatile Denoising-Based Approximate Message Passing for Compressive Sensing.

Approximate message passing-based compressive sensing reconstruction has received increasing attention, the performance of which depends heavily on the ability of the denoising operator. However, most methods only employ an off-the-shelf denoising model as the denoising operator of the iteration solver, which imposes an unfavorable limit on reconstruction performance of compressive sensing. To solve the aforementioned issue, we propose a novel versatile denoising-based approximate message passing model, abbreviated as VD-AMP, for compressive sensing (CS) recovery. To be specific, we meticulously design a double encoder-decoder denoising network (DEDNet), which manifests the impressive performance in Gaussian denoising. Moreover, a fine-grained noise level division (FNLD) solution is proposed to release the potential of the well-designed DEDNet so as to improve the reconstruction performance. However, strengthening the denoiser alone fails to remove the distortion artifact of reconstruction images at low sampling rates. To alleviate the defect, we propose an anti-aliasing sampling (AS), which firstly maps the input image to a smoothing sub-space using the proposed DEDNet before vanilla sampling, reducing aliasing between high-frequency and low-frequency information on measurement. Extensive experiments on benchmark datasets demonstrate that the proposed VD-AMP significantly outperforms state-of-the-art CS reconstruction models by a large margin, e.g., up to 2dB gains on PSNR.

Nonlocal Structured Sparsity Regularization Modeling for Hyperspectral Image Denoising

The non-local-based model for hyperspectral image (HSI) denoising first uses non-local self-similarity (NSS) prior to group similar full-band patches into three-dimensional non-local full-band groups (tensors) using a block matching (BM) operation, and then a low-rank (LR) penalty is typically applied to each non-local full-band group to reduce noise. While non-local-based methods have shown promising performance in HSI denoising, most existing methods have only considered the LR property of the non-local full-band group while ignoring the strong correlation between sparse coefficients. Moreover, such methods often result in unsatisfactory visual artifacts due to the noise sensitivity of BM operations, while requiring expensive computations. To address these limitations, this paper proposes a novel non-local structured sparsity regularization (NLSSR) approach for HSI denoising. First, to mitigate the noise sensitivity of the BM operation, we propose a graph-based domain distance scheme to index similar full-band patches to form the non-local full-band group. Second, we design an adaptive unidirectional low-rank (LR) dictionary with low complexity that takes into account the differences in intrinsic structure correlation among different modes of the non-local full-band tensor. Third, we utilize a global spectral LR prior to reduce spectral redundancy. Fourth, we develop a generalized soft-thresholding (GST) algorithm based on the alternating minimization framework to solve the NLSSR-based HSI denoising problem. We perform extensive experiments on both simulated and real data to show that the proposed NLSSR algorithm outperforms many popular or state-of-the-art HSI denoising methods in both quantitative and visual evaluations.

A New Virtual Tracking Sub-Algorithm Based Hybrid Active Control System for Narrowband Noise With Impulsive Interference

Mechanical noise is usually a mixture of narrowband and impulsive noise which needs complex active noise control (ANC) algorithms to improve the de-noising performance. But the ANC algorithm with a high computation load will reduce the real-time performance of an ANC system, thus decreasing the attenuation performance and even leading to divergence. To alleviate this contradiction in narrowband ANC systems, a new virtual filtered-x L0 norm discrete Fourier cancellation (FxL0DFC) based hybrid FxNLMS(filtered-x normalized least mean square)-FxDFC framework is proposed to decrease the total computing load and keep good attenuation performance. For a fast-changing noise, the FxNLMS algorithm is employed. The new virtual FxL0DFC algorithm serves to prepare parameters for steady-state, and when this happens, the FxDFC algorithm with the parameters provided by FxL0DFC is applied. Compared to using the FxNLMS algorithm to attenuate narrowband periodical noise, the FxDFC algorithm has nearly the same tracking performance while having a low computational load. As a result, the FxL0DFC-based FxNLMS-FxDFC algorithm leads to a reduction of the total computational load. Moreover, the proposed method performs excellently in terms of tracking in simulations and experiments on actual data, particularly in environments with rapid power changes and impulsive noise.

Attention based deep convolutional U-Net with CSA optimization for hyperspectral image denoising

Hyperspectral image (HSI) de-noising plays a significant role in HSI quality enhancement because it consists of rich image information. Although, large amounts of information commonly have a lot of noise which will greatly impact HSI processing and application. Deep convolution neural networks (CNNs) have been predominant in image de-noising because of recent advances in deep learning. However, CNNs may result in performance degradation due to plain network depth growth. It leads to face challenges in the shallow image feature fusion and limits the potential of the network to extract HSI features. To overcome these problems, an attention-based deep convolutional U-Net is proposed in this research work. It improves the down-sampling and up-sampling strategies to extract the image features and restore the HSI information, respectively. In order to provide better de-noising performance, the edge information of the image is extracted with the aid of a depthwise and polarized self-attention mechanism. Moreover, the clonal selection optimization algorithm (CSOA) minimizes the loss function obtained from the proposed system by selecting optimal parameters. The dataset used to evaluate the proposed model is the Real-World Hyperspectral Images database and is implemented in the working platform of Python. The experimental result shows that the proposed system performs better than existing state-of-the-art de-noising schemes. The achieved PSNR, SSIM, SAM and ERGAS performance metrics values are: 35.82, 0.94, 3.14 and 19.42, respectively.

Compound fault diagnosis for industrial robots based on dual-transformer networks

The accurate diagnosis of the compound fault of industrial robots can be highly beneficial to maintenance management. In the actual noisy working environment of industrial robots, the mixed and feeble failure features are easy to be overwhelmed, which poses a major challenge for the industrial robot compound fault diagnosis. Meanwhile, in the existing studies, a large-size deep learning model is the guarantee of decent denoising and fault diagnosis performance. However, this demands expensive computational costs and large data samples, which are not always available. In order to address both challenges, in this study, an integrated approach that contains two compact Transformer networks is proposed to achieve accurate compound fault diagnosis for industrial robots. In this approach, the feedback current signals collected from a six-axis industrial robot are first transformed into time-frequency image representation via continuous wavelet transformation (CWT). Secondly, a novel deep learning algorithm called compact Uformer is proposed to denoise the time-frequency image. Subsequently, the denoised time-frequency images are fed into compact convolutional Transformer (CCT) for compound fault diagnosis. An experimental study based on a real-world industrial robot compound fault dataset was conducted. The experimental results reveal that the proposed method can achieve satisfactory compound fault diagnosis accuracy based on the data collected from the noisy environment in comparison with the state-of-the-art algorithms.