Reliable Samples Research Articles

Source-free domain adaptation framework for fault diagnosis of rotation machinery under data privacy

Domain adaptation-based fault diagnosis method can learn knowledge from the labeled data in the related source rotation machinery and achieve reliable performance in the unlabeled target machinery. However, due to the impact of data privacy, the labeled source data are usually unavailable, severely restricting the application of the domain adaptation-based fault diagnosis methods. In this paper, a source-free domain adaptation framework is proposed to decouple the direct usage of the source data. First, the adaptive prototype memory matrix is constructed to select reliable samples, which are employed to define the pseudo labels. Then, the confidence-based filter is designed to improve the reliability of pseudo labels. Finally, the curriculum learning strategy is utilized to balance the optimal weight between the source information and target data. The effectiveness of the proposed method was verified by datasets in the rig and real wind turbine gearboxes.

Active consistency network for multi-source domain generalization in brain tumor segmentation

Recent advances in deep learning for brain tumor segmentation demonstrate good performance when the training and test data share the same distribution. However, medical images from different medical institutions often face distribution shifts in clinical applications. To deal with domain shift in multi-source domain generalization, we propose Active Consistency Network (ACN) with a single encoder and dual decoders to learn domain invariant features. A causal feature learning strategy and uncertainty-based refinery strategy are proposed to obtain reliable soft labels of reliable samples (Active Consistency Labels, ACL). Active consistency labels remove the noisy and unreliable samples for model consistency. Causality-based Random Masking (CRM) is proposed to obtain more stable causal features by adversarial learning. Causal features help achieve stable uncertainty estimation. Besides, two decoders are supervised by the active consistency labels learned from each other’s decoders, which can help decoders learn domain invariant features. Experimental results show that ACN utilizes causal feature learning and active consistency labels improving segmentation performance on multi-institutional brain tumor segmentation and achieving state-of-the-art performance compared to other domain generalization methods.

The VVV near-IR galaxy catalogue in a Northern part of the Galactic disc

ABSTRACT The automated identification of extragalactic objects in large surveys provides reliable and reproducible samples of galaxies in less time than procedures involving human interaction. However, regions near the Galactic disc are more challenging due to the dust extinction. We present the methodology for the automatic classification of galaxies and non-galaxies at low Galactic latitude regions using both images and photometric and morphological near-IR data from the VISTA Variables in the Vía Láctea eXtended (VVVX) survey. Using the VVV NIR Galaxy Catalogue (VVV NIRGC), we analyse by statistical methods the most relevant features for galaxy identification. This catalogue was used to train a convolutional neural network with image data and an XGBoost model with both photometric and morphological data and then to generate a data set of extragalactic candidates. This allows us to derive probability catalogues used to analyse the completeness and purity as a function of the configuration parameters and to explore the best combinations of the models. As a test case, we apply this methodology to the Northern disc region of the VVVX survey, obtaining 172 396 extragalactic candidates with probabilities of being galaxies. We analyse the performance of our methodology in the VVV disc, reaching an F1-score of 0.67, a 65 per cent purity, and a 69 per cent completeness. We present the VVV NIRGC: Northern part of the Galactic disc comprising 1003 new galaxies, with probabilities greater than 0.6 for either model, with visual inspection and with only two previously identified galaxies. In the future, we intend to apply this methodology to other areas of the VVVX survey.

Unsupervised domain adaptation via reliable pseudolabeling based memory module and dynamic distance threshold learning

An important challenge in unsupervised domain adaptation (UDA) is how to capture domain-specific discriminative information in the target domain data. For the labels in the target domain are unavailable, some methods uncover the intrinsic discriminative information by clustering the target domain features in the embedding space. However, most of existing works only consider filtering reliable target domain samples by setting a fixed distance threshold, ignoring the potential intraclass and interclass relationships and cannot set appropriate distance thresholds for different transfer tasks and different classes. To this end, we propose a new pseudolabeling framework named the dynamic distance threshold learning framework (DDT) for UDA. In our framework, we construct distance memory module to encode intraclass and interclass distribution information, and select reliable target domain samples satisfying intraclass compactness and interclass independence via dynamic intraclass and interclass distance thresholds. Specifically, we define class centroid memory modules, intraclass and interclass distance memory modules in the embedding space. Then, we retrieve the intraclass compactness distance threshold and interclass independence distance threshold from the distance memory based on the class and classpair information of target domain sample. Finally, we use reliable target domain samples to update the class centroid and the distance memory module, so that the distance memory module is dynamically updated following the model training. Benefiting from DDT, the UDA model can be updated with reliable target domain samples for training, and the joint distribution can be further aligned at the category level. As a lightweight technique, DDT constructs class center and distance memory modules to transfer the intraclass and interclass distribution relationships from the source domain to the target domain, and sets dynamic intraclass and interclass distances for different classes in different transfer tasks. Without any additional training parameters, the framework can be easily plugged into various domain adaptation paradigms. Experiments on benchmark datasets show that our method can filter the noise in pseudolabels and confers a competitive advantage compared with existing state-of-the-art domain adaptation methods.

An evolution strategy of GAN for the generation of high impedance fault samples based on Reptile algorithm

In a distribution system, sparse reliable samples and inconsistent fault characteristics always appear in the dataset of neural network fault detection models because of high impedance fault (HIF) and system structural changes. In this paper, we present an algorithm called Generative Adversarial Networks (GAN) based on the Reptile Algorithm (GANRA) for generating fault data and propose an evolution strategy based on GANRA to assist the fault detection of neural networks. First, the GANRA generates enough high-quality analogous fault data to solve a shortage of realistic fault data for the fault detection model’s training. Second, an evolution strategy is proposed to help the GANRA improve the fault detection neural network’s accuracy and generalization by searching for GAN’s initial parameters. Finally, Convolutional Neural Network (CNN) is considered as the identification fault model in simulation experiments to verify the validity of the evolution strategy and the GANRA under the HIF environment. The results show that the GANRA can optimize the initial parameters of GAN and effectively reduce the calculation time, the sample size, and the number of learning iterations needed for dataset generation in the new grid structures.

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.

Improved learning performance for small datasets in high dimensions by new dual-net model for non-linear interpolation virtual sample generation

The number of reliable samples obtained in early decision-making activity is usually relatively small. Due to variable distribution and incomplete structure of tiny datasets, it is challenging to create reliable and robust predictive modeling using classic statistical and machine learning models in small sample settings. The virtual sample generation (VSG) technique improves model learning accuracies for minimal datasets across diverse applications. Virtual samples on independent variables were generated using established VSG methods predicated on the assumption of a probability distribution or a membership function to fill data gaps. However, in the actual world, non-linear function interactions between variables are common. To address this issue, this paper developed a novel VSG method called Dual-VSG, which generates non-linear interpolation virtual samples using a self-supervised learning (SSL) framework to improve learning performance on small datasets. We generated non-linear interpolation virtual samples without labels by estimating non-linear functions and transforming them into a high-dimensional space using the proposed dual-net model. The weights of the dual-net model are transferred to a downstream task to generate virtual sample labels. To demonstrate the effectiveness of the suggested strategy, this research employed five datasets. On the Backpropagation Neural Networks (BPNN) predictive model, we compared the suggested method's prediction performance to two state-of-the-art VSG approaches. To assess prediction performance on a regression dataset, the Mean Absolute Percentage Error (MAPE) and the Root Mean Square Error (RMSE) are used. Furthermore, the classification accuracy (ACC) and the Fl measure are used to assess classification capability on classification datasets. In addition, the paired t-test was utilized to see if the suggested Dual-VSG approach differed significantly from the other VSG methods in terms of RMSE, MAPE, accuracy (ACC), or F1 score. For short datasets, the suggested Dual-VSG method outperforms those VSG methods, according to our experimental results.

RAIH-Det: An End-to-End Rotated Aircraft and Aircraft Head Detector Based on ConvNeXt and Cyclical Focal Loss in Optical Remote Sensing Images

In airport ground-traffic surveillance systems, the detection of an aircraft and its head (AIH) is an important task in aircraft trajectory judgment. However, accurately detecting an AIH in high-resolution optical remote sensing images is a challenging task due to the difficulty in effectively modeling the features of aircraft objects, such as changes in appearance, large-scale differences, complex compositions, and cluttered background. In this paper, we propose an end-to-end rotated aircraft and aircraft head detector (RAIH-Det) based on ConvNeXt-T (Tiny) and cyclical local loss. Firstly, a new U-shaped network based on ConvNeXt-T with the same performance as the Local Vision Transformer (e.g., Swin Transformer) is presented to assess the relationships among aircraft in the spatial domain. Then, in order to enhance the sharing of more mutual information, the extended BBAVectors with six vectors captures the oriented bounding box (OBB) of the aircraft in any direction, which can assist in head keypoint detection by exploiting the relationship between the local and overall structural information of aircraft. Simultaneously, variant cyclical focal loss is adopted to regress the heatmap location of keypoints on the aircraft head to focus on more reliable samples. Furthermore, to perform a study on AIH detection and simplify aircraft head detection, the OBBs of the “plane” category in the DOTA-v1.5 dataset and the corresponding head keypoints annotated by our volunteers were integrated into a new dataset called DOTA-Plane. Compared with other state-of-the-art rotated object and keypoint detectors, RAIH-Det, as evaluated on DOTA-Plane, offered superior performance.

Entropy-Optimized Fault Diagnosis Based on Unsupervised Domain Adaptation

In practice, the cross-domain transfer of data distribution and the sample imbalance of fault status are inevitable, but one or both are often ignored, which restricts the adaptability and classification accuracy of the generated fault diagnosis (FD) model. Accordingly, an entropy-optimized method is proposed in this paper based on an unsupervised domain-adaptive technique to enhance FD model training. For the training, pseudosamples and labels corresponding to the target samples are generated through data augmentation and self-training strategies to diminish the distribution discrepancy between the source and target domains. Meanwhile, an adaptive conditional entropy loss function is developed to improve the data quality of the semisupervised learning, with which reliable samples are generated for the training. According to the experiment results, compared with other state-of-the-art algorithms, our method can achieve significant accuracy improvement in rolling bearing FD. Typically, the accuracy improvement compared with the baseline Convolutional Neural Network (CNN) is achieved by over 13.23%.

Deep-sea freezer

Recovery of samples from the deep ocean in pristine condition is difficult due to large environmental differences between the deep and surface waters through which the samples necessarily must be transported. Here, we propose a concept for deep-sea sample recovery: a deep-sea freezer using thermoelectric cooling capable of generating ice in the deep and recover them frozen on-board ships. As a proof of concept, we present the DSF-α, a prototype Deep-Sea Freezer based on Peltier device rated at 2000 m. In situ assessments of the DSF-α on remotely operated vehicles showed its capacity to reach freezing (-13.0°C) temperatures in the deep, as well as recovering seawater frozen on deck. Although the DSF-α is limited in that achieving sufficient freezing for useful sample recovery is time consuming, the deep-sea freezer opens a whole frontier of new possibilities for preserving various types of deep-sea samples and has the potential to be adapted according to various needs of the deep-sea research community. With the first literal ‘marine snow’ in the deep, we offer a glimpse to a future where the recovery of reliable bathyal samples is no longer laborious.

Hybrid AI model for power transformer assessment using imbalanced DGA datasets

AbstractArtificial intelligence (AI) methods have been used widely in power transformer fault diagnosis with notable developments in solutions for big data problems. Training data is essential to accurately train AI models. The volume, scope and variety of data samples contribute significantly to the success and reliability of diagnostic outcomes. This paper provides a comprehensive review and comparison of different augmentation methods used to generate reliable data samples for minority and majority classes to balance the diversity and distribution of dissolved gas analysis (DGA) datasets. The augmentation method presented in this paper combines three common AI models—the Support Vector Machine (SVM), Decision Tree, and k‐Nearest Neighbour (KNN)—to assess performance for diagnostic fault determination and classification, with comparator assessment using no data augmentation. Comparative analysis of the hybrid models uses evaluation metrics including accuracy, precision, recall, specificity, F‐score, G‐mean, and the area under receiver operation characteristic (Auc). Experimental results presented in this paper confirm that the data augmentation applied to AI models can resolve difficulties in imbalanced data distribution and provide significant improvements for fault diagnosis, particularly for minority classes.

Hierarchical memory-guided long-term tracking with meta transformer inquiry network

Long-term tracking is a critical and rapidly developing field in visual object tracking research. The tracking target’s appearance can change over time due to its motion or environmental factors, resulting in various appearance patterns. Those factors, such as dense distractors, confusing backgrounds, and motion blurs, make it difficult to track objects. Existing tracking algorithms typically rely on online learning to adapt to such long-term variations. However, obtaining reliable training samples and implementing effective updating schemes can be difficult, particularly when the target object frequently disappears and reappears. Therefore, in this paper, we propose a Hierarchical Memory-guided Long-term Tracker (HMLT) with a Meta Transformer Inquiry Network (MTIN) that refines online learning. Our method introduces a hierarchical memory strategy that considers simple and trustworthy updating of long-term tracking components using multiple target memories. We also devise MTIN to identify available memory based on the long-term variation pattern of the target, so as to avoid the risk of updating from incorrect samples. In addition, we devise a memory attention network to perform robust redetection based on long-term memory. Based on the hierarchical memory strategy, we construct a complete and learnable long-term tracking framework that uses a validator learned from memory to reconcile local and global searches. Our experimental results on several benchmarks, including LaSOT, VOT-LT2018, VOT-LT2019, TLP, OTB-2015, and UAV123, demonstrate that our proposed method achieves comparable performance to the state-of-the-art long-term tracking algorithms.

A Novel End-to-End Unsupervised Change Detection Method with Self-Adaptive Superpixel Segmentation for SAR Images

Change detection (CD) methods using synthetic aperture radar (SAR) data have received significant attention in the field of remote sensing Earth observation, which mainly involves knowledge-driven and data-driven approaches. Knowledge-driven CD methods are based on the physical theoretical models with strong interpretability, but they lack the robust features of being deeply mined. In contrast, data-driven CD methods can extract deep features, but require abundant training samples, which are difficult to obtain for SAR data. To address these limitations, an end-to-end unsupervised CD network based on self-adaptive superpixel segmentation is proposed. Firstly, reliable training samples were selected using an unsupervised pre-task. Then, the superpixel generation and Siamese CD network were integrated into the unified framework to train them end-to-end until the global optimal parameters were obtained. Moreover, the backpropagation of the joint loss function promoted the adaptive adjustment of the superpixel. Finally, the binary change map was obtained. Several public SAR CD datasets were used to verify the effectiveness of the proposed method. The transfer learning experiment was implemented to further explore the ability to detect the changes and generalization performance of our network. The experimental results demonstrate that our proposed method achieved the most competitive results, outperforming seven other advanced deep-learning-based CD methods. Specifically, our method achieved the highest accuracy in OA, F1-score, and Kappa, and also showed superiority in suppressing speckle noise, refining change boundaries, and improving detection accuracy in a small area change.

A joint local–global search mechanism for long-term tracking with dynamic memory network

Long-term tracking is a very popular tracking framework recently. Previous trackers mostly adopted a sliding window approach for global search, which can lead to significant time consumption. In addition, frequent appearance variations can make the tracking model lack long term adaptability. To solve the above issues, this paper proposes a joint local–global search mechanism for long-term tracking with dynamic memory network (LGST), which uses a global detector to generate “high-quality” region proposals and a dynamic memory network to store reliable template samples. Specifically, we first propose a novel global re-detection module that employs learning target channel-aware representation (TCA-R) to guide the region proposal network (RPN) to perform global search and cover the region of the target of interest to cope with the target disappearance and reappearance. Then, for the local tracker, a dynamic memory network is designed to collect the updated templates at different time periods and store reliable template samples. More importantly, we connect and feed the collected template features into the transformer encoder to generate the attention-enhanced features. Finally, we adopt a verifier to determine whether the tracked object exists or not, and dynamically switch between local search and global search mechanisms in the next image frame to form a long-term tracking framework. Experimental results on VOT-LT2018, VOT-LT2019, LaSOT, TLP, OTB2015, and UAV123 benchmarks show that our proposed tracker achieves comparable performance to state-of-the-art tracking algorithms.

AttentionDTA: Drug-Target Binding Affinity Prediction by Sequence-Based Deep Learning With Attention Mechanism.

The identification of drug-target relations (DTRs) is substantial in drug development. A large number of methods treat DTRs as drug-target interactions (DTIs), a binary classification problem. The main drawback of these methods are the lack of reliable negative samples and the absence of many important aspects of DTR, including their dose dependence and quantitative affinities. With increasing number of publications of drug-protein binding affinity data recently, DTRs prediction can be viewed as a regression problem of drug-target affinities (DTAs) which reflects how tightly the drug binds to the target and can present more detailed and specific information than DTIs. The growth of affinity data enables the use of deep learning architectures, which have been shown to be among the state-of-the-art methods in binding affinity prediction. Although relatively effective, due to the black-box nature of deep learning, these models are less biologically interpretable. In this study, we proposed a deep learning-based model, named AttentionDTA, which uses attention mechanism to predict DTAs. Different from the models using 3D structures of drug-target complexes or graph representation of drugs and proteins, the novelty of our work is to use attention mechanism to focus on key subsequences which are important in drug and protein sequences when predicting its affinity. We use two separate one-dimensional Convolution Neural Networks (1D-CNNs) to extract the semantic information of drug's SMILES string and protein's amino acid sequence. Furthermore, a two-side multi-head attention mechanism is developed and embedded to our model to explore the relationship between drug features and protein features. We evaluate our model on three established DTA benchmark datasets, Davis, Metz, and KIBA. AttentionDTA outperforms the state-of-the-art deep learning methods under different evaluation metrics. The results show that the attention-based model can effectively extract protein features related to drug information and drug features related to protein information to better predict drug target affinities. It is worth mentioning that we test our model on IC50 dataset, which provides the binding sites between drugs and proteins, to evaluate the ability of our model to locate binding sites. Finally, we visualize the attention weight to demonstrate the biological significance of the model. The source code of AttentionDTA can be downloaded from https://github.com/zhaoqichang/AttentionDTA_TCBB.

Supervised poststack 3D seismic data classification via multiscale and multilabel consistent feature reduction

Following the advancement of machine learning-based seismic feature classification techniques for complex reservoirs, the acquisition and analysis of reliable seismic samples involved in seismic facies analysis and network-based inversion have emerged as a current research hotspot in the field of intelligent seismic processing. Many investigations focus on the improvement of model classification algorithms and neural networks. However, creating and collecting labels for massive seismic data are highly time-consuming and laborious, and suffer from sample unreliability and category imbalance in the case of small-sample labels. To address such problems, a multiscale and multilabel consistent principal component analysis-linear discriminant analysis (PCA-LDA) algorithm to learn a robust feature discriminative dictionary for classification is presented. In addition to the automatic use of multilabels from well logs and core analysis, we have associated multiscale with well trajectory locations to enrich sample information and enhance the reliability of the samples during 3D sample acquisition. More specifically, we begin by proposing an approach for the automatic collection of multiscale multilabel 3D poststack seismic samples along the well track. Next, the multilabel sequence in the scan window is fed into the Boyer-Moore majority vote algorithm for sample segmentation, which constructs multilabel hierarchies for each sample. Then to enhance the model training bias due to small-sample label imbalance, we develop a novel label-shuffling balanced strategy, which obtains a complete database by filling random unduplicated augmented training samples (spatial and frequency-domain augmentation operations). Finally, the linear robustness decision-making space of PCA-LDA is obtained using the feature mapping space of PCA, as well as its visual representation. Experimental results on synthetic and field seismic data demonstrate that robust feature extraction with a trustworthy and complete multiscale and multilabel sample database increases classification accuracy.

Point Cloud Instance Segmentation with Inaccurate Bounding-Box Annotations.

Most existing point cloud instance segmentation methods require accurate and dense point-level annotations, which are extremely laborious to collect. While incomplete and inexact supervision has been exploited to reduce labeling efforts, inaccurate supervision remains under-explored. This kind of supervision is almost inevitable in practice, especially in complex 3D point clouds, and it severely degrades the generalization performance of deep networks. To this end, we propose the first weakly supervised point cloud instance segmentation framework with inaccurate box-level labels. A novel self-distillation architecture is presented to boost the generalization ability while leveraging the cheap but noisy bounding-box annotations. Specifically, we employ consistency regularization to distill self-knowledge from data perturbation and historical predictions, which prevents the deep network from overfitting the noisy labels. Moreover, we progressively select reliable samples and correct their labels based on the historical consistency. Extensive experiments on the ScanNet-v2 dataset were used to validate the effectiveness and robustness of our method in dealing with inexact and inaccurate annotations.

A fully automatic and high-accuracy surface water mapping framework on Google Earth Engine using Landsat time-series

ABSTRACT Efficient and continuous monitoring of surface water is essential for water resource management. Much effort has been devoted to the task of water mapping based on remote sensing images. However, few studies have fully considered the diverse spectral properties of water for the collection of reference samples in an automatic manner. Moreover, water area statistics are sensitive to the satellite image observation quality. This study aims to develop a fully automatic surface water mapping framework based on Google Earth Engine (GEE) with a supervised random forest classifier. A robust scheme was built to automatically construct training samples by merging the information from multi-source water occurrence products. The samples for permanent and seasonal water were mapped and collected separately, so that the supplement of seasonal samples can increase the spectral diversity of the sample space. To reduce the uncertainty of the derived water occurrences, temporal correction was applied to repair the classification maps with invalid observations. Extensive experiments showed that the proposed method can generate reliable samples and produce good-quality water mapping results. Comparative tests indicated that the proposed method produced water maps with a higher quality than the index-based detection methods, as well as the GSWD and GLAD datasets.

SURE: Screening unlabeled samples for reliable negative samples based on reinforcement learning

For many classification tasks, particularly in the bioinformatics field, only experimentally validated positive samples are available, and experimentally validated negative samples are not recorded. The lack of negative samples poses a challenge for using machine learning to perform such tasks. To address this problem, we propose a novel deep reinforcement learning-based model to screen reliable negative samples from unlabeled samples, named SURE. The model has two modules: a sample selector and a sample inspector. The sample selector screens reliable negative samples from unlabeled samples by two reinforcement strategies (learn to identify positive samples and reduce sample noise) and feeds the screened samples into the sample inspector. The sample inspector classifier provides rewards to the sample selector. The two modules are trained together to optimize the sample selector and sample inspector strategies. In this paper, we focus on one popular issue in the bioinformatics field: the ncRNA-protein interaction (NPI) prediction task, which lacks reliable negative samples. Thirty datasets for NPI prediction are used to test the screening effect of SURE. The experimental results show that our model has a robust negative sample screening capability and is superior to all outstanding sample screening methods used in the NPI prediction task. In addition, we refine 5 NPI datasets containing reliable negative samples screened by SURE, and a web server (ww.csbg-jlu.info/sure) is available offering the NPI prediction refined by SURE.

Self-supervised clustering with assistance from off-the-shelf classifier

Deep clustering outperforms conventional clustering by mutually promoting representation learning and cluster assignment. However, most existing deep clustering methods suffer from two major drawbacks. Firstly, most cluster assignment methods are highly dependent on the intermediate target distribution generated by a handcrafted nonlinear mapping function. Secondly, the clustering results can be easily guided towards wrong direction by the misassigned samples in each cluster. The existing deep clustering methods are incapable of discriminating such samples. These facts largely limit the possible performance that deep clustering methods can reach. To address these issues, a novel Self-Supervised Clustering (SSC) framework is constructed, which boosts the clustering performance by classification in an unsupervised manner. Fuzzy theory is used to score the membership of each sample to the clusters in terms of probability in each training epoch, which evaluates the intermediate clustering result certainty of each sample. The most reliable samples can be selected with the help of a sample selection method according to the membership and enhanced by data augmentation method. These augmented data are employed to fine-tune an off-the-shelf deep network classifier with the labels provided by the clustering in a self-supervised way. The classification results of the original dataset are used as the target distribution to guide the training process of the deep clustering model. The proposed framework can efficiently discriminate sample outliers and generate better target distribution with the assistance of the powerful classifier. Extensive experiments indicate that the proposed framework remarkably outperforms state-of-the-art deep clustering methods on four benchmark datasets.