Sample Labeling Research Articles

Improving the crop classification performance by unlabeled remote sensing data

Accurate and timely crop classification results play a crucial role in providing data support for agricultural policy-making and crop yield estimation. However, the current development of crop classification faces a bottleneck in improving classification performance due to limited labeled samples and saturated classification algorithms. In this study, we propose a novel method to improve crop classification performance by leveraging unlabeled remote sensing data (URSD). Importantly, our method does not necessitate a large number of labeled samples or significant modifications to the classification algorithm. Instead, it relies on a unique self-supervised training approach and a substantial amount of URSD. Specifically, we develop a self-supervised classification framework based on a Multilayer Perceptron (MLP) and introduce a self-supervised training approach that takes into account both temporal and spectral factors. Additionally, we construct a historical sample classification model based on crop growth knowledge, emphasizing the correlation of local time series. We evaluate the proposed method using four study areas in China. The analysis of pre-training data types reveals that our method not only improves the classification performance of current year samples but also demonstrates noticeable improvement in classifying historical samples. The classification method analysis demonstrates the ability of our proposed self-supervised learning training approach to accumulate more prior knowledge. Overall, these results highlight the advantages of our method in terms of classification efficiency and performance improvement.

Near-Optimal Active Learning for Multilingual Grapheme-to-Phoneme Conversion

The construction of pronunciation dictionaries relies on high-quality and extensive training data in data-driven way. However, the manual annotation of corpus for this purpose is both costly and time consuming, especially for low-resource languages that lack sufficient data and resources. A multilingual pronunciation dictionary includes some common phonemes or phonetic units, which means that these phonemes or units have similarities in the pronunciation of different languages and can be used in the construction process of pronunciation dictionaries for low-resource languages. By using a multilingual pronunciation dictionary, knowledge can be shared among different languages, thus improving the quality and accuracy of pronunciation dictionaries for low-resource languages. In this paper, we propose using shared articulatory features among multiple languages to construct a universal phoneme set, which is then used to label words for multiple languages. To achieve this, we first developed a grapheme−phoneme (G2P) model based on an encoder−decoder deep neural network. We then adopted a near-optimal active learning method in the process of building the pronunciation dictionary to select informative samples from a large, unlabeled corpus and had them labeled by experts. Our experiments demonstrate that this method selected about 1/5 of the unlabeled data and achieved an even higher conversion accuracy than the results of the large data training method. By selectively labeling samples with a high uncertainty in the model, while avoiding labeling samples that were accurately predicted by the current model, our method greatly enhances the efficiency of pronunciation dictionary construction.

Small samples-oriented intrinsically explainable machine learning using Variational Bayesian Logistic Regression: An intensive care unit readmission prediction case for liver transplantation patients

Recently, machine learning has revolutionized medical diagnosis and prediction, rivaling human experts in accuracy. However, the limited interpretability poses challenges in meeting stringent medical decision standards. To facilitate interpretable predictive inference, this article proposes a small samples-oriented intrinsically explainable machine learning model, an Optimal Variational Bayesian Logistic Regression (OVBLR) model with a salient feature estimation strategy (OVBLR-SFE). OVBLR-SFE sequentially learns from limited labeled samples, mining disease-related risk factors to provide intrinsic interpretability to its predictive inference automatically. The model achieves this by reasonably approximating the posterior probability through the proposed OVBLR model to estimate the feature weight and influential effect on model outputs to facilitate salient feature selection. The effectiveness and superiority of OVBLR-SFE are demonstrated through extensive experiments on four benchmark medical datasets from the UCI repository, including the Bone Marrow Transplant: Children Data Set, Breast Cancer Wisconsin (Original) Data Set, SPECT Heart Data Set, and Heart Disease Data Set, as well as a real-world case of intensive care unit readmission prediction. Experimental results reveal that the OVBLR-SFE can achieve high accuracies of 92.02%, 97.21%, 86.71%, and 84.47%, respectively, with an average classification accuracy of 90.10% stably on the four UCI datasets. Especially, OVBLR-SFE can accurately infer significant features that align with that obtained by the post-hoc interpretation approaches: SHAP and AcME, demonstrating remarkable interpretability. In the real application, the proposed method achieved a readmission prediction accuracy of 88.11% in predicting the readmission probabilities of liver transplantation patients. Notably, the inferred salient features, such as total Intensive Care Unit length of stay (ICUTime), N-terminal pro-brain natriuretic peptide(NTproBNP), Operating time of surgery(OperationTime_h), and intraoperative red blood cell transfusion (RBCT), emerges as crucial in readmission prediction, aligning with the assessments of clinical experts. The source code of this study can be accessed at https://github.com/xqw42/OVBLR-SFE.

Cellular Network Fault Diagnosis Method Based on a Graph Convolutional Neural Network.

The efficient and accurate diagnosis of faults in cellular networks is crucial for ensuring smooth and uninterrupted communication services. In this paper, we propose an improved 4G/5G network fault diagnosis with a few effective labeled samples. Our solution is a heterogeneous wireless network fault diagnosis algorithm based on Graph Convolutional Neural Network (GCN). First, the common failure types of 4G/5G networks are analyzed, and then the graph structure is constructed with the data in the network parameter, given data sets as nodes and similarities as edges. GCN is used to extract features from the graph data, complete the classification task for nodes, and finally predict the fault types of cells. A large number of experiments are carried out based on the real data set, which is achieved by driving tests. The results show that, compared with a variety of traditional algorithms, the proposed method can effectively improve the performance of network fault diagnosis with a small number of labeled samples.

National-Standards- and Deep-Learning-Oriented Raster and Vector Benchmark Dataset (RVBD) for Land-Use/Land-Cover Mapping in the Yangtze River Basin

A high-quality remote sensing interpretation dataset has become crucial for driving an intelligent model, i.e., deep learning (DL), to produce land-use/land-cover (LULC) products. The existing remote sensing datasets face the following issues: the current studies (1) lack object-oriented fine-grained information; (2) they cannot meet national standards; (3) they lack field surveys for labeling samples; and (4) they cannot serve for geographic engineering application directly. To address these gaps, the national-standards- and DL-oriented raster and vector benchmark dataset (RVBD) is the first to be established to map LULC for conducting soil water erosion assessment (SWEA). RVBD has the following significant innovation and contributions: (1) it is the first second-level object- and DL-oriented dataset with raster and vector data for LULC mapping; (2) its classification system conforms to the national industry standards of the Ministry of Water Resources of the People’s Republic of China; (3) it has high-quality LULC interpretation accuracy assisted by field surveys rather than indoor visual interpretation; and (4) it could be applied to serve for SWEA. Our dataset is constructed as follows: (1) spatio-temporal-spectrum information is utilized to perform automatic vectorization and label LULC attributes conforming to the national standards; and (2) several remarkable DL networks (DenseNet161, HorNet, EfficientNetB7, Vision Transformer, and Swin Transformer) are chosen as the baselines to train our dataset, and five evaluation metrics are chosen to perform quantitative evaluation. Experimental results verify the reliability and effectiveness of RVBD. Each chosen network achieves a minimum overall accuracy of 0.81 and a minimum Kappa of 0.80, and Vision Transformer achieves the best classification performance with overall accuracy of 0.87 and Kappa of 0.86. It indicates that RVBD is a significant benchmark, which could lay a foundation for intelligent interpretation of relevant geographic research about SWEA in the Yangtze River Basin and promote artificial intelligence technology to enrich geographical theories and methods.

Application of big data technology in electromechanical operation and maintenance intelligent platform

Abstract Aiming at the data preprocessing requirements and label data cost issues arising from the intelligent operation and maintenance of electromechanical equipment, this article mainly studies structured data cleaning methods and fault prediction algorithms for a small number of label samples. First, this article introduces the overall architecture of the intelligent operation and maintenance system for electromechanical equipment. Second, based on the electromechanical equipment operation and maintenance data access service, data cleaning, and fault prediction, this article constructs an electromechanical equipment intelligent operation and maintenance platform based on Kafka message queue, Spark cluster, and other components, and introduces the functional composition of the system in detail. Finally, the article describes the functions of each component of data access service, data cleaning, and fault prediction in detail. To address the cost issue associated with sufficient labeled sample data for data analysis, we propose a semi-supervised learning algorithm, IF-GBDT, based on improved independent forests and Gradient Boosting Decision Tree. The independent forest algorithm supplements labels for unlabeled data based on the learning results of a small number of labeled samples. We also use the gradient lifting tree algorithm to train the model based on the new tag data set for fault prediction, thereby reducing the impact of lack of tags on the accuracy of the prediction model. Experiments show that this method improves classification accuracy and has good adaptability and concurrency performance for a small number of tags.

Prior knowledge-based residuals shrinkage prototype networks for cross-domain fault diagnosis

In engineering practice, device failure samples are limited in the case of unexpected catastrophic faults, thereby limiting the application of deep learning in fault diagnosis. In this study, we propose a prior knowledge-based residual shrinkage prototype network to resolve the fault diagnosis challenges under limited labeled samples. First, our method combines general supervised learning and metric meta-learning to extract prior knowledge from the labeled source data by utilizing a denoised residual shrinkage network. Further, the knowledge extracted from the supervised learning is used for prototype metric training to achieve a better feature representation for the fault diagnosis. Finally, our approach outperforms a series of baseline methods in the few-shot cross-domain diagnostic task on the gearbox and bearing datasets. A diagnosis accuracy of more than 95% has been achieved in a variety of working conditions for diagnostic tasks, which is far higher than the existing basic method.

IS3L: An Integrated Self-training Semi-Supervised Learning Strategy for Underwater Acoustic Target Detection

This paper proposed IS3L (Integrated Self-training Semi-Supervised Learning), a universal underwater acoustic target detection strategy with a semi-supervised learning architecture, which can comprehensively refine weakly labeled samples and label samples automatically to boost the detection accuracy. To our best knowledge, this is the first work to combine semi-supervised learning with underwater target detection. A universal feature extraction strategy is employed to extract comprehensive features ranging from time/frequent domain to dedicated auditory parameters. An integrated self-training strategy is proposed to generate pseudo labels from unlabeled samples and meanwhile avoid wrong labeling via a voting mechanism. Then, the final detection can be made by the rear classifier, which has been fully trained with original labeled samples and generated pseudo-labeled samples. Experimental results, which are based on real sea trail data, show that the proposed algorithm can reach up to 98.13% in detection accuracy with only 50 labeled samples (7040 samples in total).

Network Collaborative Pruning Method for Hyperspectral Image Classification Based on Evolutionary Multi-Task Optimization

Neural network models for hyperspectral images classification are complex and therefore difficult to deploy directly onto mobile platforms. Neural network model compression methods can effectively optimize the storage space and inference time of the model while maintaining the accuracy. Although automated pruning methods can avoid designing pruning rules, they face the problem of search efficiency when optimizing complex networks. In this paper, a network collaborative pruning method is proposed for hyperspectral image classification based on evolutionary multi-task optimization. The proposed method allows classification networks to perform the model pruning task on multiple hyperspectral images simultaneously. Knowledge (the important local sparse structure of the network) is automatically searched and updated by using knowledge transfer between different tasks. The self-adaptive knowledge transfer strategy based on historical information and dormancy mechanism is designed to avoid possible negative transfer and unnecessary consumption of computing resources. The pruned networks can achieve high classification accuracy on hyperspectral data with limited labeled samples. Experiments on multiple hyperspectral images show that the proposed method can effectively realize the compression of the network model and the classification of hyperspectral images.

Nanopore Filter: A Method for Counting and Extracting Single DNA Molecules Using a Biological Nanopore.

This paper describes a method for the real-time counting and extraction of DNA molecules at the single-molecule level by nanopore technology. As a powerful tool for electrochemical single-molecule detection, nanopore technology eliminates the need for labeling or partitioning sample solutions at the femtoliter level. Here, we attempt to develop a DNA filtering system utilizing an α-hemolysin (αHL) nanopore. This system comprises two droplets, one filling with and one emptying DNA molecules, separated by a planar lipid bilayer containing αHL nanopores. The translocation of DNA through the nanopores is observed by measuring the channel current, and the number of translocated molecules can also be verified by quantitative polymerase chain reaction (qPCR). However, we found that the issue of contamination seems to be an almost insolvable problem in single-molecule counting. To tackle this problem, we tried to optimize the experimental environment, reduce the volume of solution containing the target molecule, and use the PCR clamp method. Although further efforts are still needed to achieve a single-molecule filter with electrical counting, our proposed method shows a linear relationship between the electrical counting and qPCR estimation of the number of DNA molecules.

Domain-adaptive graph neural network for few-shot learning

The task of few-shot classification (FSC) is to build a model to discriminate novel categories that do not present in training categories with limited labeled samples. However, in many applications, the training categories and novel categories are assumed to come from the same domain. Existing few-shot classification algorithms achieve promising performance in a single domain, but often fail to generalize to unseen domains because of the different feature distribution across domains. The main goal of this work is to propose an effective recognition model that can work on various image domains with domain shift. Specifically, we propose to construct a new domain-attention mechanism with some adapters based on the squeeze-and-excitation network architectures. The proposed network processes multiple domains simultaneously and all parameters are shared across domains. In addition, because of the large discrepancy in feature distributions among different domains, similarity transformation layers are used to reduce the differences in the feature distributions in the training stage. Extensive experiments were conducted to validate the domain generalization capability (GC) of this model on four FSC datasets: mini-ImageNet, Cars, CUB-200-2011, and Simpsons Characters Data. The results show that the presented method has excellent performances on various datasets across diverse domains.

A Semisupervised Method With Swarm Intelligence Optimization for Intelligent Fault Diagnosis

With the recent development of deep learning, machine learning-based methods have gained promising achievements for fault diagnosis. However, most of these methods are supervised and a large amount of labeled data is required for training in practical applications. To deal with the problem of limited labeled samples in fault diagnosis tasks, a semi-supervised deep learning-based method, named particle swarm joint classifier with an auto-encoder, is proposed in this paper. In this method, a classifier and an auto-encoder are developed using labeled data and unlabeled data respectively, and two loss functions are combined to train these two models simultaneously. In addition, an optimization strategy based on the greedy algorithm and particle swarm optimization is designed and applied to optimize the combined weights of the loss function. The proposed method is verified experimentally on two popular rotating machinery datasets: the Case Western Reserve University bearing dataset and the Paderborn University bearing dataset. The experimental results have demonstrated that the proposed method could achieve a classification accuracy of over 95% on these two datasets with no more than 20 labeled samples per class, and the proposed optimization strategy could improve the classification accuracy significantly when reducing the number of parameters.

Guidance on modeling circulating miRNA to distinguish multiple cancer types by an observation of large-scale open data.

e13537 Background: Cell-free miRNAs (cf-miRNA), circulated in body fluids such as plasma or serum, have shown their ability to detect, diagnose, and monitor cancers. Combining machine learning (ML) technology with these biomarkers facilitates early detection of cancers, which increases the accuracy of clinical decisions and empowers people to take control of their health status. However, the data of the cf-miRNAs has characteristics, which will affect the results of ML. Therefore, this study tries to expound on them in different aspects and to build a reasonable model. Methods: We downloaded large-scale datasets of the platform GPL21263 from the Gene Expression Omnibus for modeling experiments. We curated 8,174 subjects with 2,565 miRNA targets across 7 cancer types of different cf-miRNA-based cancer studies. Moreover, we used principal component analysis (PCA) to observe the datasets, recursive feature elimination (RFE) for feature selection, and tree-based algorithms to build the prediction model. Results: The characteristics of the cf-miRNA we like to share are: (1) Cancer subjects express more cf-miRNAs than control subjects. In the control group, there were 294 and 327 miRNAs with missing rates under 50% and 25%, respectively. In contrast, there were 395 and 485 miRNAs with the same thresholds in the cancer group; (2) Dividing subjects into cancer and control groups is simpler than distinguishing specific cancer types. In the PCA, the average Euclidean distance between the control group and each cancer type is 98.48, while it is 20.23 within each cancer type; (3) For obtaining cancer-specific biomarkers, we suggested that other non-target cancer subjects should be considered as negative controls. We modeled 7 cancer types and compared the proportion of cancer-specific biomarkers, not selected by any other models. The proportion increased from 30.0% to 57.8% after we added other non-target cancer subjects to the control group; next, we focus on multi-cancer modeling: (4) We need at least 400 samples to distinguish seven cancer types. In our experiment, we kept increasing the size of the training data, a hundred at a time. As data was added to the model, the accuracy increased but plateaued after adding 400 samples; (5) Based on RFE, 120 miRNAs is a reasonable number to distinguish seven cancers. Moreover, we found some of these miRNAs are only expressed in cancer subjects. We might lose this kind of biomarkers if we filtered out them by the missing rate; (6) The 10-fold cross validation accuracy of the multi-cancer model can achieve 93.0% using the gradient-boosted trees algorithm. Conclusions: In this study, we showed the guidance for modeling miRNAs in different aspects, including labeling strategy, sample and feature sizes, and the high-accuracy multi-cancer model we can achieve. We hope this guidance would inspire researchers on cf-miRNA-related machine learning applications.

Semi-supervised classifier ensemble model for high-dimensional data

To complete the challenging task of high-dimensional data classification with limited labeled samples, we propose two semi-supervised learning models, namely the random subspace classifier ensemble model (SSRS) and its adaptive version (ASSRS). Considering the unique characteristics of high-dimensional data, SSRS selects subspaces of sample and feature dimensions and then reduces the dimensions of each subspace. To improve SSRS performance further, we designed a sample-labeling auxiliary algorithm, adaptive sample subspace algorithm, and adaptive weight voting rule for ASSRS to increase the proportion of labeled samples, obtain a suitable sample subspace for each feature subspace, and acquire a relative optimal weight for each base classifier. Experiments revealed that the performances of SSRS and ASSRS were better than those of other competitive algorithms and that the performance of ASSRS was stronger than that of SSRS. Additionally, we can accurately label samples in datasets where the proportion of labeled samples is relatively low by using SSRS and ASSRS. Because analysts are facing large numbers of high-dimensional datasets with limited labels, it is important to make accurate predictions based on a limited proportion of labeled data.

Hyperspectral Image Classification Based on Multiscale Hybrid Networks and Attention Mechanisms

Hyperspectral image (HSI) classification is one of the most crucial tasks in remote sensing processing. The attention mechanism is preferable to a convolutional neural network (CNN), due to its superior ability to express information during HSI processing. Recently, numerous methods combining CNNs and attention mechanisms have been applied in HSI classification. However, it remains a challenge to achieve high-accuracy classification by fully extracting effective features from HSIs under the conditions of limited labeled samples. In this paper, we design a novel HSI classification network based on multiscale hybrid networks and attention mechanisms. The network consists of three subnetworks: a spectral-spatial feature extraction network, a spatial inverted pyramid network, and a classification network, which are employed to extract spectral-spatial features, to extract spatial features, and to obtain classification results, respectively. The multiscale fusion network and attention mechanisms complement each other by capturing local and global features separately. In the spatial pyramid network, multiscale spaces are formed through down-sampling, which can reduce redundant information while retaining important information. The structure helps the network better capture spatial features at different scales, and to improve classification accuracy. Experimental results on various public HSI datasets demonstrate that the designed network is extremely competitive compared to current advanced approaches, under the condition of insufficient samples.

Transfer learning for cross-scene 3D pavement crack detection based on enhanced deep edge features

Inspired by heterogeneity of rapid-increasing 3D pavement data and the generalization ability of transfer learning, a robust and generalized framework for cross-scene 3D pavement-crack detection and attribute extraction was proposed in this paper, called profile component decomposition model with holistically nested edge detection (PCDM-HED). The core purpose of the PCDM-HED is to construct the enhanced deep edge features. By applying profile frequency, sparse characteristics of 3D profiles, and fusing the multiscale and multilevel edge characteristics of 3D depth maps in HED network, the robust enhanced feature can highlight the essential properties of cracks in heterogeneous 3D data. It overcomes the complex domain shifts caused by different 3D imaging conditions, pavement textures, and crack distributions in heterogeneous data. Cross-domain transfer experiments were carried out over seven 3D/2D datasets with 915 pavement sections. The results show that proposed PCDM-HED achieved average buffered Hausdorff scores of 90.17 to 96.42, recall scores of 0.84 to 0.91, and F-values of 0.85 to 0.89 in six different datasets without labeled samples. Compared with 9 groups of comparison results, including the traditional and related state-of-the-art methods, the transfer generalization effect of proposed PCDM-HED is more than 23% higher than that of comparison results. The proposed PCDM-HED makes full use of limited off-the-shelf samples, demonstrated strong transfer learning capability. It provides an effective solution for heterogeneous 3D pavement-crack detection tasks in engineering, in the case of limited labeled samples or even no corresponding labeled samples.

A deep learning framework for early diagnosis of Alzheimer’s disease on MRI images

Numerous medical studies have shown that Alzheimer’s disease (AD) was present decades before the clinical diagnosis of dementia. As a result of the development of these studies with the discovery of many ideal biomarkers of symptoms of Alzheimer’s disease, it became clear that early diagnosis requires a high-performance computational tool to handle such large amounts of data, as early diagnosis of Alzheimer’s disease provides us with a healthy opportunity to benefit from treatment. The main objective of this paper is to establish a complete framework that is based on deep learning approaches and convolutional neural networks (CNN). Four stages of AD, such as (I) preprocessing and data preparation, (II) data augmentation, (III) cross-validation, and (IV) classification and feature extraction based on deep learning for medical image classification, are implemented. In these stages, two methods are implemented. The first method uses a simple CNN architecture. In the second method, the VGG16 model is the pre-trained model that is trained on the ImageNet dataset but applies the same model to the different datasets. We apply transfer learning, meaning, and fine-tuning to take advantage of the pre-trained models. Seven performance metrics are used to evaluate and compare the two methods. Compared to the most recent effort, the proposed method is proficient of analyzing AD, moreover, entails less labeled training samples and minimal domain prior knowledge. A significant performance gain on classification of all diagnosis groups was achieved in our experiments. The experimental findings demonstrate that the suggested designs are appropriate for basic structures with minimal computational complexity, overfitting, memory consumption, and temporal regulation. Besides, they achieve a promising accuracy, 99.95% and 99.99% for the proposed CNN model in the classification of the AD stage. The VGG16 pre-trained model is fine-tuned and achieved an accuracy of 97.44% for AD stage classifications.

A Mask Self-Supervised Learning-Based Transformer for Bearing Fault Diagnosis With Limited Labeled Samples

In recent years, Transformer has become an effective tool for fault diagnosis, but it has been shown that a sufficient amount of labeled data is usually required to train a Transformer model. However, a few labeled data can be obtained in the actual industrial process, and labeling a large quantity of training samples is costly. To reduce the demand for training labeled samples, this paper proposes a mask self-supervised learning-based Transformer (MSFormer) for bearing fault diagnosis of multistage centrifugal fans in petrochemical units under the condition of limited samples. In mask self-supervised learning, unlabeled samples can be used to mine robust representations of fault signals and potential relationships between sub-sequences to obtain a pre-trained model with well-generalized parameters. Then, a few labeled samples are utilized to fine-tune by supervised learning to enable MSFormer the discrimination ability to identify different bearing fault types. The effectiveness of the proposed method is fully validated on the multistage centrifugal fan dataset and the Case Western Reserve University motor bearing dataset. The experimental results demonstrate that MSFormer is effective in reducing the number of labeled training samples, and compared to state-of-the-art methods, MSFormer has superior diagnosis performance under the condition of limited labeled samples.

GIGAN: Self‐supervised GAN for generating the invisible using cycle transformation and conditional normalization

AbstractObjects in a real scene often occlude each other and inferring a complete appearance from the visible part is an important and challenging task. In this paper, the authors propose a self‐supervised generative adversarial network GIGAN (GAN for generating the invisible), which can generate the complete appearance of objects without labelled invisible part information. The authors build two cycle transformation networks CycleIncomplete (CycleI) and CycleComplete (CycleC) that share parameters to improve the accuracy of mask completion. This design does not require well‐matched training images and can make better use of the limited labelled samples. In addition, the authors propose a conditional normalization module and combine it with the inferred complete mask output. The combination not only enhances the content recovery ability and obtains more realistic outputs, but also improves the efficiency of the generation process. Experimental results show that compared with existing self‐supervised learning models, our method achieves l1 error, mean intersection‐over‐union (mIOU), and Fréchet inception distance (FID) improvements on the COCOA and KINS datasets.

Gearbox compound fault diagnosis method based on deep adversarial graph convolution transfer learning network under low label ratios

Advancement in measurement techniques has dramatically contributed to the development of the modern manufacturing industry. As the primary fault causing unplanned downtime of mechanical equipment, gearbox compound faults are usually coupled by single faults with unequal severity and are difficult to obtain. In industrial scenarios, monitoring data for extreme operating conditions is not available in advance, and labeling samples is time-consuming and costly. However, current research on unseen compound fault diagnosis relies on numerous labeled samples to train the model, and few studies are conducted on transfer learning and unseen compound fault diagnosis under low label ratios. To address the issue, a gearbox compound fault diagnosis method based on deep adversarial graph convolution transfer learning network (DAGCTLN) under low label ratios is proposed. Specifically, a novel DAGCTLN model, including a feature extractor, two label classifiers, and a discriminator, is constructed to realize the diagnosis of faults in the transfer domain and unseen compound faults in the source and target domains. The feature extractor of a three-layer graph convolutional network is presented to achieve deep extraction of fault features under low label ratios. Then a domain space adversarial mechanism between the feature extractor and discriminator is used to achieve global alignment of the source and target domain features. Furthermore, two label classifiers are constructed, and the adversarial adaptation of the decision boundary is realized by maxi-min the classifier difference to achieve subdomain alignment of the same class features in all domains. Experimental results indicate that DAGCTLN can achieve the highest fault diagnosis accuracy in the transfer domain compared to state-of-the-art algorithms. The average diagnosis accuracy of compound faults in all domains can reach 98.41% even if the label ratio is only 0.1, which provides guiding significance for the safe operation and predictive maintenance of mechanical equipment.