Unsupervised Domain Adaptation Research Articles

A source free robust domain adaptation approach with pseudo-labels uncertainty estimation for rolling bearing fault diagnosis under limited sample conditions

As essential components of machinery equipment, rolling bearings directly affect the safety of the machinery equipment. The timely diagnosis of bearing faults can effectively prevent equipment lapses. However, bearings are often inconsistently distributed. This has resulted in a significant decrease in their availability. Moreover, the performances of traditional models are poor when fault samples are scarce. The unsupervised domain adaptation (UDA) model based on the transfer learning theory can solve the above problems in static scenarios. However, source domain data are often not directly accessible for privacy protection. Therefore, achieving the robustness of UDA models is significantly challenging. Source-free UDA can achieve a positive transfer from the source domain to the target domain based only on a pretrained source-domain model and unlabeled target-domain data. In this study, we built a source-free robust UDA approach with pseudo-label uncertainty estimation (SFRDA-PLUE) for diagnosing bearing faults using a limited number of samples. First, we designed a robust feature extractor (SANet) and proposed a novel binary soft-constrained information entropy. This was applied to solve the problem that standard information entropy cannot effectively estimate the uncertainty of pseudo-labels. In addition, we constructed a weighted comparison filter strategy to smoothen the fuzzy samples. Finally, we introduced an information-maximizing loss strategy to optimize the performance of the source domain classifier and the pseudo-label estimator. Thus, the robustness of the pseudo-label uncertainty estimation was significantly improved. The experimental results validated that the SFRDA-PLUE approach can achieve excellent diagnostic performance under a limited number of samples.

Enhancing quantitative 1H NMR model generalizability on honey from different years through partial least squares subspace and optimal transport based unsupervised domain adaptation

Honey is a nourishing and natural food product that is widely favored by a diverse group of consumers. Proton Nuclear Magnetic Resonance (1H NMR) is a powerful tool for quantitative analysis of honey and plays a crucial role in ensuring its quality. The 1H NMR technique necessitates the utilization of multivariate calibration models to facilitate the quantitative analysis of key compounds present in honey. However, maintaining consistent measurement conditions across different years is scarcely possible, which can significantly impact the distribution of training and test spectra, ultimately leading to reduced performance of predictive models. Unsupervised domain adaptation (UDA) methods have gained considerable attention for their ability to match distribution differences between the labeled source spectra and the unlabeled target spectra without costly annotation. To enhance the quantitative model generalizability on honey from different years, we propose a UDA method known as partial least squares subspace and optimal transport-based UDA (PLSS-OT-UDA). This approach eliminates distribution differences between the source subspace and target subspace via partial least squares (PLS) dimensionality reduction and OT. Firstly, the optimal latent variable weight matrix from the source domain (i.e., labeled 1H NMR data in 2017) is extracted with PLS. Next, the dimension of both source and target domains (i.e., unlabeled 1H NMR data in 2018) is reduced and their corresponding subspaces are obtained with weight matrix of the source domain. Finally, OT is then employed to align the distribution of the source and target domains within the subspace. Experimental results on the honey dataset demonstrate that the PLSS-OT-UDA outperforms traditional methods, including transfer component analysis (TCA), optimal transport for domain adaptation (OTDA), domain adaptation based on principal component analysis and optimal transport (PCA-OTDA), and subspace alignment (SA), with respect to generalization performance on three components: baume degree, sugar content, and water content.

AVATAR: Adversarial self-superVised domain Adaptation network for TARget domain

This paper introduces AVATAR (adversarial self-supervised domain adaptation network for target domain), a novel unsupervised domain adaptation method designed to address the challenge of transferring knowledge from a well-labeled source domain to an unlabeled target domain. AVATAR’s key contributions lie in its unique combination of adversarial learning and self-supervised learning techniques, which enhance the model’s adaptability and accuracy in new, unlabeled environments. The technical innovations of AVATAR include a domain adversarial learning component that reduces domain discrepancy by aligning source and target domain features in a shared feature space, a self-supervised learning module that leverages pseudo-labels and instance weights to refine target domain discrimination boundaries, and a sample selection strategy guided by within-cluster distances to focus on aligning informative samples while mitigating the impact of outliers. By integrating these components, AVATAR provides a comprehensive framework for tackling complex unsupervised domain adaptation tasks, particularly when the domain gap is significant. The method balances domain-invariant feature learning with domain-specific adjustments, enabling effective adaptation to dynamically changing environments. Extensive experiments on multiple benchmark datasets demonstrate AVATAR’s superior performance compared to existing domain adaptation methods, especially in scenarios with limited labeled target domain data. These results highlight AVATAR’s potential as a versatile and robust solution for domain adaptation challenges in various real-world applications.

In the Wild Video Violence Detection: An Unsupervised Domain Adaptation Approach

This work addresses the challenge of video violence detection in data-scarce scenarios, focusing on bridging the domain gap that often hinders the performance of deep learning models when applied to unseen domains. We present a novel unsupervised domain adaptation (UDA) scheme designed to effectively mitigate this gap by combining supervised learning in the train (source) domain with unlabeled test (target) data. We employ single-image classification and multiple instance learning (MIL) to select frames with the highest classification scores, and, upon this, we exploit UDA techniques to adapt the model to unlabeled target domains. We perform an extensive experimental evaluation, using general-context data as the source domain and target domain datasets collected in specific environments, such as violent/non-violent actions in hockey matches and public transport. The results demonstrate that our UDA pipeline substantially enhances model performances, improving their generalization capabilities in novel scenarios without requiring additional labeled data.

Time-series domain adaptation via sparse associative structure alignment: Learning invariance and variance

Domain adaptation on time-series data, which is often encountered in the field of industry, like anomaly detection and sensor data forecasting, but received limited attention in academia, is an important but challenging task in real-world scenarios. Most of the existing methods for time-series data use the covariate shift assumption for non-time-series data to extract the domain-invariant representation, but this assumption is hard to meet in practice due to the complex dependence among variables and a small change of the time lags may lead to a huge change of future values. To address this challenge, we leverage the stableness of causal structures among different domains. To further avoid the strong assumptions in causal discovery like linear non-Gaussian assumption, we relax it to mine the stable sparse associative structures instead of discovering the causal structures directly. Besides the domain-invariant structures, we also find that some domain-specific information like the strengths of the structures is important for prediction. Based on the aforementioned intuition, we extend the sparse associative structure alignment model in the conference version to the Sparse Associative Structure Alignment model with domain-specific information enhancement (SASA2 in short), which aligns the invariant unweighted spare associative structures and considers the variant information for time-series unsupervised domain adaptation. Specifically, we first generate the segment set to exclude the obstacle of offsets. Second, we extract the unweighted sparse associative structures via sparse attention mechanisms. Third, we extract the domain-specific information via an autoregressive module. Finally, we employ a unidirectional alignment restriction to guide the transformation from the source to the target. Moreover, we further provide a generalization analysis to show the theoretical superiority of our method. Compared with existing methods, our method yields state-of-the-art performance, with a 5% relative improvement in three real-world datasets, covering different applications: air quality, in-hospital healthcare, and anomaly detection. Furthermore, visualization results of sparse associative structures illustrate what knowledge can be transferred, boosting the transparency and interpretability of our method.

A bidirectional domain separation adversarial network based transfer learning method for near-infrared spectra

Traditional calibration transfer (CT) methods usually fail to adapt the source domain model to the target domain because of changes associated with the instrument, detection environment, sample composition, or sample type. Most deep transfer learning (DTL) methods have shown reliable capability in extracting domain-invariant representations. However, factors unique to the spectral domain limit their development and application, including difficulty in obtaining labeled samples, numerous spectral perturbation factors, and the scarcity of spectral databases. To address these challenges, we here introduce the Bidirectional Domain-Separating Adversarial Network (BiDSAN), based on a deep adversarial network transfer architecture, for learning and adapting to distribution/domain differences between sample spectra. BiDSAN incorporates Wasserstein divergence (w-div) and dynamic adversarial factor into the domain critic (discriminator) to accelerate the extraction of spectral shared components. Additionally, the use of the spectral signal distortion ratio as a loss function and the introduction of a bidirectional mechanism further enhanced the training accuracy and speed of spectral models. Calibration analyses were tested on pharmaceutical and two sets of wood datasets, corresponding to three common domain/distribution difference scenarios discussed in the study. Compared to existing CT and DTL methods, the proposed BiDSAN method demonstrated superior spectral signal calibration capability and stability, achieving unsupervised domain adaptation for similar samples and, for the first time, obtaining ideal prediction results in cross-species spectral domain adaptation. Our method represents significant potential for using DTL approaches in spectral domain adaptation.

Cross-Domain Object Detection through Consistent and Contrastive Teacher with Fourier Transform

The teacher–student framework has been employed in unsupervised domain adaptation, which transfers knowledge learned from a labeled source domain to an unlabeled target domain. However, this framework suffers from two serious challenges: the domain gap, causing performance degradation, and noisy teacher pseudo-labels, which tend to mislead students. In this paper, we propose a Consistent and Contrastive Teacher with Fourier Transform (CCTF) method to address these challenges for high-performance cross-domain object detection. To mitigate the negative impact of domain shifts, we use the Fourier transform to exchange the low-frequency components of the source and target domain images, replacing the source domain inputs with the transformed image, thereby reducing domain gaps. In addition, we encourage the localization and classification branches of the teacher to make consistent predictions to minimize the noise in the generated pseudo-labels. Finally, contrastive learning is employed to resist the impact of residual noise in pseudo-labels. After extensive experiments, we show that our method achieves the best performance. For example, our model outperforms previous methods by 3.0% on FoggyCityscapes.

ACAN: A plug-and-play Adaptive Center-Aligned Network for unsupervised domain adaptation

Domain adaptation is an important topic due to its capability in transferring knowledge from source domain to target domain. However, many existing domain adaptation methods primarily concentrate on aligning the data distributions between the source and target domains, often neglecting discriminative feature learning. As a result, target samples with low confidence are embedded near the decision boundary, where they are susceptible to being misclassified, resulting in negative transfer. To address this problem, a novel Adaptive Center-Aligned Network dubbed ACAN is proposed for unsupervised domain adaptation in this work. The main innovations of ACAN are fourfold. Firstly, it is a plug-and-play module and can be easily incorporated into any domain alignment methods without increasing the model complexity and computational burden. Secondly, in contrast to conventional softmax plus cross-entropy loss, angular margin loss is called to enhance the discrimination power for classifier. Thirdly, entropy regularization is exploited to highlight the probability of potential related class, which renders our learned feature representation far away from the decision boundary. Fourthly, to improve the discriminative capacity of model to the target domain, we propose to align the target domain samples to the corresponding class center via pseudo labels. Incorporating ACAN, the performance of baseline domain alignment methods is significantly improved. Extensive ablation and comparison experiments on four widely adopted databases demonstrate the effectiveness of our ACAN. Code is available at: https://github.com/Cloudfly-Z/ACAN

ADAC: an active domain adaptive network with progressive learning strategy for cloud detection of remote sensing imagery

ABSTRACT To enhance the adaptability and application capability of the cloud detection model in different remote sensing satellite domains, unsupervised domain adaptation methods are employed to improve the model’s robustness and transferability. However, conducting fully unlabeled training in the target domain to be adapted may impose difficulty on further improving the accuracy. It is more practical to obtain a few labeled target data for domain adaptation. Additionally, the detection of complex cloud scenes remains a persistent challenge. To address these issues, an Active Domain Adaptation network for Cloud detection (ADAC) is proposed in this paper. This network employs active learning to select a small number of representative samples for annotation, facilitating the transfer of a source domain-trained model to the target domain. Moreover, a novel Generator (G) designed for cloud segmentation is developed, where G utilizes a progressive learning strategy by dividing the network into two stages to learn gradient information, guiding cloud boundary segmentation and enabling the distinction between cloud-like objects and clouds. Furthermore, G integrates Transformer and Convolutional Neural Network to address the issues of misclassifying thin clouds. The proposed method is validated on Landsat 8 and GF-2 datasets, representing the source and target domains, respectively. The experimental results demonstrate ADAC achieved a 4% improvement in the overall accuracy with only 5% labeled training samples in the target domain, achieving the effective adaptation performance in the GF-2 domain.

Single-source unsupervised domain adaptation for cross-subject MI-EEG classification based on discriminative information

Single-source unsupervised domain adaptation for cross-subject MI-EEG classification based on discriminative information

Dual-drive RUL prediction of gear transmission systems based on dynamic model and unsupervised domain adaption under zero sample

Gear transmission systems are key components in rotating machinery, and its remaining useful life (RUL) prediction can provide adequate leading time for well-timed maintenance. Existing RUL prediction models usually rely on sufficient training data. However, the acquisition of full life cycle dataset of gear transmission system is difficult or impossible. To solve the problem on the difficult startup and poor generalization of prediction model under zero environment, a dual-drive prediction method based on dynamic model of gear transmission system and unsupervised domain adaptation is proposed in this paper. Firstly, a dynamic model of gear transmission system and growth mechanism of local defect is established to generate full-life cycle simulation data. Then, the multi-scale modulation features are extracted based on simulated and measured data. Furthermore, multi-scale temporal convolution operations are introduced into dual-channel unsupervised domain adaptation model. Besides, a compound principle of reverse truncation and forward expansion principle is investigated to determine the first prediction time. Finally, the validity of the proposed model is verified by two kinds of gearbox data. Ablation experiments are carried out to evaluate the contribution of each module in proposed model. In addition, the effectiveness and generalization ability of proposed model are verified when compared with other advanced transfer learning methods.

Kernel Extreme Learning Machine with Discriminative Transfer Feature and Instance Selection for Unsupervised Domain Adaptation

The goal of domain adaptation (DA) is to develop a robust decision model on the source domain effectively generalize to the target domain data. State-of-the-art domain adaptation methods typically focus on finding an optimal inter-domain invariant feature representation or helpful instances from the source domain. In this paper, we propose a kernel extreme learning machine with discriminative transfer features and instance selection (KELM-DTF-IS) for unsupervised domain adaptation tasks, which consists of two steps: discriminative transfer feature extraction and classification with instance selection. At the feature extraction stage, we extend cross domain mean approximation(CDMA) by incorporating a penalty term and develop discriminative cross domain mean approximation (d-CDMA) to optimize the category separability between instances. Subsequently, d-CDMA is integrated into kernel ELM-AutoEncoder(KELM-AE) for extracting inter-domain invariant features. During the classification process, our approach uses CDMA metrics to compute a weights to each source instances based on their impact in reducing distribution differences between domains. Instances with a greater effect receive higher weights and vice versa. These weights are then used to distinguish and select source domain instances before incorporating them into weight KELM for proposing an adaptive classifier. Finally, we apply our approach to conduct classification experiments on publicly available domain adaptation datasets, and the results demonstrate its superiority over KELM and numerous other domain adaptation approaches.

Cross-Modality Medical Image Segmentation via Enhanced Feature Alignment and Cross Pseudo Supervision Learning.

Given the diversity of medical images, traditional image segmentation models face the issue of domain shift. Unsupervised domain adaptation (UDA) methods have emerged as a pivotal strategy for cross modality analysis. These methods typically utilize generative adversarial networks (GANs) for both image-level and feature-level domain adaptation through the transformation and reconstruction of images, assuming the features between domains are well-aligned. However, this assumption falters with significant gaps between different medical image modalities, such as MRI and CT. These gaps hinder the effective training of segmentation networks with cross-modality images and can lead to misleading training guidance and instability. To address these challenges, this paper introduces a novel approach comprising a cross-modality feature alignment sub-network and a cross pseudo supervised dual-stream segmentation sub-network. These components work together to bridge domain discrepancies more effectively and ensure a stable training environment. The feature alignment sub-network is designed for the bidirectional alignment of features between the source and target domains, incorporating a self-attention module to aid in learning structurally consistent and relevant information. The segmentation sub-network leverages an enhanced cross-pseudo-supervised loss to harmonize the output of the two segmentation networks, assessing pseudo-distances between domains to improve the pseudo-label quality and thus enhancing the overall learning efficiency of the framework. This method's success is demonstrated by notable advancements in segmentation precision across target domains for abdomen and brain tasks.

Progressive Decision Boundary Shifting for Unsupervised Domain Adaptation.

Unsupervised domain adaptation (UDA) is attracting more attention from researchers for boosting the task-specific generalization on target domain. It focuses on addressing the domain shift between the labeled source domain and the unlabeled target domain. Recent biclassifier-based UDA models perform category-level alignment to reduce domain shift, and meanwhile, self-training is used for improving the discriminability of target instances. However, the error accumulation problem of instances with high semantic uncertainty may cause discriminability degradation and category-level misalignment. To solve this issue, we design the progressive decision boundary shifting algorithm, where stable category information of target instances is explored for learning a discriminability structure on target domain. Specifically, we first model the semantic uncertainty of instances by progressively shifting decision boundaries of category. Then, we introduce the uncertainty decoupling in a contrastive manner, where the discriminative information is learned from the source domain for instance with low semantic uncertainty. Furthermore, we minimize the predictive entropy of instances with high semantic uncertainty to reduce their prediction confidence. Extensive experiments on three popular datasets show that our model outperforms the current state-of-the-art (SOTA) UDA methods.

Semantic-aware Message Broadcasting for Efficient Unsupervised Domain Adaptation.

Vision transformer has demonstrated great potential in abundant vision tasks. However, it also inevitably suffers from poor generalization capability when the distribution shift occurs in testing (i.e., out-of-distribution data). To mitigate this issue, we propose a novel method, Semantic-aware Message Broadcasting (SAMB), which enables more informative and flexible feature alignment for unsupervised domain adaptation (UDA). Particularly, we study the attention module in the vision transformer and notice that the alignment space using one global class token lacks enough flexibility, where it interacts information with all image tokens in the same manner but ignores the rich semantics of different regions. In this paper, we aim to improve the richness of the alignment features by enabling semantic-aware adaptive message broadcasting. Particularly, we introduce a group of learned group tokens as nodes to aggregate the global information from all image tokens, but encourage different group tokens to adaptively focus on the message broadcasting to different semantic regions. In this way, our message broadcasting encourages the group tokens to learn more informative and diverse information for effective domain alignment. Moreover, we systematically study the effects of adversarial-based feature alignment (ADA) and pseudo-label based self-training (PST) on UDA. We find that one simple two-stage training strategy with the cooperation of ADA and PST can further improve the adaptation capability of the vision transformer. Extensive experiments on DomainNet, OfficeHome, and VisDA-2017 demonstrate the effectiveness of our methods for UDA.

PSO-based unified framework for unsupervised domain adaptation in image classification

PSO-based unified framework for unsupervised domain adaptation in image classification

Cross-domain fault diagnosis network based on attributes and features transfer with dual classifier under limited and unbalanced datasets

Abstract Deep learning (DL)-based methods have demonstrated significant success in fault diagnosis owing to their robust feature extraction and non-linear fitting capabilities. Meanwhile, their remarkable performance is accompanied by constant operating conditions and sufficient monitoring data. However, in real engineering environments, variable working conditions or limited and unbalanced data are common, which can widen the gap between fault diagnosis methods and real industrial applications. In this paper, we proposed a cross-domain fault diagnosis network based on a dual classifier (CFDNet) with input being limited and unbalanced data to learn attributes and features for unsupervised domain adaptation. We found that the diagnostic performance is commonly bounded by the underlying knowledge, especially feature extraction from original data. Therefore, we designed a new feature encoder with features and relationships, i.e. using a convolutional neural network and graph convolutional network, which improves extraction efficiency while retaining valuable information. Then, we discovered that enforced feature transfer can lead to negative transfer. To mitigate this, we present a feature and attribute transfer framework, which not only achieves features transfer but also enables attributes transfer. Furthermore, it was noted that limited and unbalanced datasets can introduce label bias and lead to biased model training. Hence, we designed dual classifiers to improve the probability of high-confidence final prediction by synthesizing diagnostic results. Comprehensive experiments conducted on three case studies demonstrate the effectiveness and superiority of our method for cross-domain fault diagnosis under limited and unbalanced datasets, which outperforms state-of-the-art methods in this study.

Unsupervised Domain Adaptation for Inter-Session Re-Calibration of Ultrasound-Based HMIs.

Human-Machine Interfaces (HMIs) have gained popularity as they allow for an effortless and natural interaction between the user and the machine by processing information gathered from a single or multiple sensing modalities and transcribing user intentions to the desired actions. Their operability depends on frequent periodic re-calibration using newly acquired data due to their adaptation needs in dynamic environments, where test-time data continuously change in unforeseen ways, a cause that significantly contributes to their abandonment and remains unexplored by the Ultrasound-based (US-based) HMI community. In this work, we conduct a thorough investigation of Unsupervised Domain Adaptation (UDA) algorithms for the re-calibration of US-based HMIs during within-day sessions, which utilize unlabeled data for re-calibration. Our experimentation led us to the proposal of a CNN-based architecture for simultaneous wrist rotation angle and finger gesture prediction that achieves comparable performance with the state-of-the-art while featuring 87.92% less trainable parameters. According to our findings, DANN (a Domain-Adversarial training algorithm), with proper initialization, offers an average 24.99% classification accuracy performance enhancement when compared to no re-calibration setting. However, our results suggest that in cases where the experimental setup and the UDA configuration may differ, observed enhancements would be rather small or even unnoticeable.

DACA: A domain adaptive fault diagnosis approach with class-aware based on cross-domain extreme imbalance data

Existing research on unsupervised domain adaptation (UDA) has primarily centered on mitigating differences in feature distribution across various domains. However, in real-world scenarios, natural data often exhibits not only a disparity in instance counts across different classes within the same domain but also across different domains, leading to imbalances at both the class and domain levels. To address this issue, this study introduces a novel scenario, termed extreme imbalance domain adaptation (EIDA), which encompasses both covariate shifts and labeling shifts. We propose a Domain Adaptation with Class-aware (DACA) framework for bearing fault diagnosis that aims to tackle inter-domain feature shifts as well as intra- and inter-domain category imbalances. Specifically, DACA enhances the recognition accuracy of minority categories without degrading the performance concerning the majority category, reduces the adverse effects of data imbalance during domain transitions, and strengthens the robustness of decision boundaries for improved spatial matching. This is achieved through four sub-modules: adversarial clustering learning, cross-domain category alignment, category-level reweighting, and category-level re-margining. The efficacy of DACA is demonstrated through simulations of three imbalance protocols in the EIDA scenario across two bearing datasets and one gearbox dataset, showcasing significant advancements in cross-domain diagnostic tasks.

LUIE: Learnable physical model-guided underwater image enhancement with bi-directional unsupervised domain adaptation

Recently, learning-based underwater enhancement (UIE) methods have made considerable progress, significantly benefiting downstream tasks such as underwater semantic segmentation and underwater depth estimation. Most existing unsupervised UIE methods utilize the atmospheric image formation model to decompose underwater images into background color, transmission map, and scene radiance. However, they rely on simplified physical models for estimating the transmission map, over-simplifying its complex formation, which results in imprecise modeling of underwater scattering effects. Additionally, supervised UIE methods heavily depend on synthetic data or ground truth, leading to limited generalization capabilities due to the substantial domain gap presented in different underwater scenarios. To tackle these challenges, we propose a Learnable physical model-guided unsupervised domain adaptation framework for Underwater Image Enhancement, dubbed LUIE. LUIE learns to predict background light, depth, and scene radiance from an underwater image. We incorporate a learnable network to estimate the transmission map based on the predicted depth map. To minimize the inter-domain gap between synthetic and real underwater images, we introduce a bi-directional domain adaptation method that alternates the background light from each domain. Experimental results demonstrate the effectiveness of our proposed method compared to existing approaches, and high-level experiment results validate that our enhanced underwater results. Experiments in real-world settings on underwater ROVs platform with NVIDIA Jetson AGX Xavier further confirm the effectiveness and efficiency of our work.