Limited Labeled Samples Research Articles

Advancing Incremental Few-Shot Video Action Recognition with Cluster Compression and Generative Separation

Few-Shot Class Incremental Learning (FSCIL) is a trending topic in deep learning, addressing the need for models to incrementally learn novel classes, particularly in real-world scenarios where continuously emerging classes come with limited labeled samples. However, the majority of FSCIL research has been dedicated to image classification and object recognition tasks, with limited attention given to video action classification. In this paper, we present a new Cluster Compression and Generative Separation (CCGS) method for Incremental Few-Shot Video Action Recognition (iFSVAR), which introduces contrastive learning to boost the degree of class separation in the base session. Simultaneously, it creates numerous fine-grained classes with diverse semantics, effectively filling the unallocated representation space. Experimental results on UCF101, Kinetics, and Something-Something-V2[Formula: see text]demonstrate the effectiveness of the framework.

Gradient-guided channel masking for cross-domain few-shot learning

Cross-Domain Few-Shot Learning (CD-FSL) addresses the Few-Shot Learning with a domain gap between source and target domains, which facilitates the transfer of knowledge from a source domain to a target domain with limited labeled samples. Current approaches often incorporate an auxiliary target dataset containing a few labeled samples to enhance model generalization on specific target domains. However, we observe that many models retain a substantial number of channels that learn source-specific knowledge and extract features that perform adequately on the source domain but generalize poorly to the target domain. This often results in compromised performance due to the influence of source-specific knowledge. To address this challenge, we introduce a novel framework, Gradient-Guided Channel Masking (GGCM), designed for CD-FSL to mitigate model channels from acquiring too much source-specific knowledge. GGCM quantifies each channel’s contribution to solving target tasks using gradients of target loss and identifies those with smaller gradients as source-specific. These channels are then masked during the forward propagation of source features to mitigate the learning of source-specific knowledge. Conversely, GGCM mutes non-source-specific channels during the forward propagation of target features, forcing the model to depend on the source-specific channels and thereby enhancing their generalizability. Moreover, we propose a consistency loss that aligns the predictions made by source-specific channels with those made by the entire model. This approach further enhances the generalizability of these channels by enabling them to learn from the generalizable knowledge contained in other non-source-specific channels. Validated across multiple CD-FSL benchmark datasets, our framework demonstrates state-of-the-art performance and effectively suppresses the learning of source-specific knowledge.

Multi-granularity self-attention mechanisms for few-shot Learning

Few-shot learning aims to classify novel data categories with limited labeled samples. Although metric-based meta-learning has shown better generalization ability as a few-shot classification method, it still faces challenges in handling data noise and maintaining inter-sample distance stability. To address these issues, our study proposes an innovative few-shot learning approach to enhance image features' global and local semantic representation. Initially, our method employs a multiscale residual module to facilitate extracting multi-granularity features within images. Subsequently, it optimizes the fusion of local and global features using the self- attention mechanism inherent in the Transformer module. Additionally, a weighted metric module is integrated to improve the model's resilience against noise interference. Empirical evaluations on CIFAR-FS and MiniImageNet few-shot datasets using 5-way 1-shot and 5-way 5-shot scenarios demonstrate the effectiveness of our approach in capturing multi-level and multi-granularity image representations. Compared to other methods, our method improves accuracy by 2.63% and 1.27% for 5-shot scenes on these two datasets. The experimental results validate the efficacy of our model in significantly enhancing few-shot image classification performance.

Teacher-Student Prototype Enhancement Network for a Few-Shot Remote Sensing Scene Classification

Few-shot remote sensing scene classification identifies new classes from limited labeled samples where the great challenges are intraclass diversity, interclass similarity, and limited supervision. To alleviate these problems, a teacher-student prototype enhancement network is proposed for a few-shot remote sensing scene classification. Instead of introducing an attentional mechanism in mainstream studies, a prototype enhancement module is recommended to adaptively select high-confidence query samples, which can enhance the support prototype representations to emphasize intraclass and interclass relationships. The construction of a few-shot teacher model generates more discriminative predictive representations with inputs from many labeled samples, thus providing a strong supervisory signal to the student model and encouraging the network to achieve accurate classification with a limited number of labeled samples. Extensive experiments of four public datasets, including NWPU-remote sens ing image scene classification (NWPU-RESISC45), aerial image dataset (AID), UC Merced, and WHU-RS19, demonstrate that this method achieves superior competitive performance than the state-of-the-art methods on five-way, one-shot, and five-shot classifications.

Advanced Global Prototypical Segmentation Framework for Few-Shot Hyperspectral Image Classification.

With the advancement of deep learning, related networks have shown strong performance for Hyperspectral Image (HSI) classification. However, these methods face two main challenges in HSI classification: (1) the inability to capture global information of HSI due to the restriction of patch input and (2) insufficient utilization of information from limited labeled samples. To overcome these challenges, we propose an Advanced Global Prototypical Segmentation (AGPS) framework. Within the AGPS framework, we design a patch-free feature extractor segmentation network (SegNet) based on a fully convolutional network (FCN), which processes the entire HSI to capture global information. To enrich the global information extracted by SegNet, we propose a Fusion of Lateral Connection (FLC) structure that fuses the low-level detailed features of the encoder output with the high-level features of the decoder output. Additionally, we propose an Atrous Spatial Pyramid Pooling-Position Attention (ASPP-PA) module to capture multi-scale spatial positional information. Finally, to explore more valuable information from limited labeled samples, we propose an advanced global prototypical representation learning strategy. Building upon the dual constraints of the global prototypical representation learning strategy, we introduce supervised contrastive learning (CL), which optimizes our network with three different constraints. The experimental results of three public datasets demonstrate that our method outperforms the existing state-of-the-art methods.

Solution for crop classification in regions with limited labeled samples: deep learning and transfer learning

ABSTRACT Reliable classification results are crucial for guiding agricultural production, forecasting crop yield, and ensuring food security. Generating reliable classification results is relatively simple in regions with sufficient labeled samples, but regions with limited labeled samples remain a significant challenge. In this study, we propose two new solutions that leverage the feature representation capabilities of deep learning and the sample reuse potential of transfer learning to solve the limited label problem. Specifically, we develop a Variable-dimensional Symmetric Network with Position Encoding (VPSNet) to improve the efficiency of labeled sample utilization. Additionally, we introduce a transfer strategy based on the Inter-Regional Discrepancies in Crop Time Series (IRDCTS) to expand the labeled sample reuse region. We evaluated the proposed model in three regions with limited labels between 2017 and 2018. Experimental results show that our model has superior discriminative feature extraction capabilities compared to other existing models. The feasibility of the proposed transfer strategy is tested in three pair regions, showing that IRDCTS can enhance the model adaptability by reducing inter-domain discrepancies. This study provides a comprehensive solution to the classification problem of the limited labeled samples, involving both the development of classification models and the implementation of transfer strategies.

Res-BiLSTMs model based on multi-task attention for real-time measurement of the free calcium oxide content

Abstract The content of free calcium oxide (f-CaO) is the primary economic index to evaluate the quality of cement. A residual bidirectional long short-term memory network model (Res-BiLSTMs) based on a multi-task attention mechanism was proposed for the characteristics of cement clinker production, used for online monitoring f-CaO content. The model utilizes the Bi-LSTM as the foundational component and combines the residual network to construct the Res-BiLSTMs coding structure, which aims to summarize the multi-level characteristic information of the input sequence. Additionally, a multi-task attention mechanism is introduced, combining the attention mechanism with semi-supervision to extract control coupling and data coupling among devices and variables. The results demonstrate that the addition of the multi-task attention mechanism led to a reduction in model errors by 0.0175 and 0.022, respectively, and an improvement in the degree of fit by 14.61%. The effectiveness of the multi-task attention mechanism for quality monitoring is confirmed. Compared to traditional LSTM, this model exhibited a reduction in errors by 0.0469 and 0.019, respectively, an increase in the correlation coefficient by 45.37%, and outperformed all other models in the comparison. The model’s measurement performance under limited labeled samples is also validated.

Cross-domain prototype similarity correction for few-shot radar modulation signal recognition

The new classes of radar signals are increasingly difficult to acquire under non-cooperative environments, which makes it difficult to support convolutional neural network training with limited labeled samples. The few-shot learning (FSL) methods have shown great performance in classification with limited labeled samples, but the FSL methods ignore that the class distributions between the new and original tasks are significantly different, resulting in a massive challenge in identifying new radar signals. To solve this problem, a few-shot radar modulation signal recognition method based on cross-domain prototype similarity correction (CDPSC) is proposed. Specifically, a residual feature tokenizer transformer (RFTT) model embedded with a pooling token generation block is designed to focus on the important features and improve the ability to represent samples. Meanwhile, the proposed domain prototype similarity mapping (DPSM) strategy adaptively learns the class mapping, reduces the inter-domain difference through feature distribution alignment, and effectively corrects the target domain prototypes. In addition, we introduce a sample prototype embedding (SPE) strategy in the training phase, which can reduce the intra-class distance and increase the inter-class distance. Experimental results demonstrate that the CDPSC method is superior to typical FSL methods in recognition accuracy under different sample numbers.

Semi-Supervised Informer for the Compound Fault Diagnosis of Industrial Robots.

The increasing deployment of industrial robots in manufacturing requires accurate fault diagnosis. Online monitoring data typically consist of a large volume of unlabeled data and a small quantity of labeled data. Conventional intelligent diagnosis methods heavily rely on supervised learning with abundant labeled data. To address this issue, this paper presents a semi-supervised Informer algorithm for fault diagnosis modeling, leveraging the Informer model's long- and short-term memory capabilities and the benefits of semi-supervised learning to handle the diagnosis of a small amount of labeled data alongside a substantial amount of unlabeled data. An experimental study is conducted using real-world industrial robot monitoring data to assess the proposed algorithm's effectiveness, demonstrating its ability to deliver accurate fault diagnosis despite limited labeled samples.

Boosting few-shot rare skin disease classification via self-supervision and distribution calibration.

Due to the difficulty in obtaining clinical samples and the high cost of labeling, rare skin diseases are characterized by data scarcity, making training deep neural networks for classification challenging. In recent years, few-shot learning has emerged as a promising solution, enabling models to recognize unseen disease classes by limited labeled samples. However, most existing methods ignored the fine-grained nature of rare skin diseases, resulting in poor performance when generalizing to highly similar classes. Moreover, the distributions learned from limited labeled data are biased, severely impairing the model's generalizability. This paper proposes a self-supervision distribution calibration network (SS-DCN) to address the above issues. Specifically, SS-DCN adopts a multi-task learning framework during pre-training. By introducing self-supervised tasks to aid in supervised learning, the model can learn more discriminative and transferable visual representations. Furthermore, SS-DCN applied an enhanced distribution calibration (EDC) strategy, which utilizes the statistics of base classes with sufficient samples to calibrate the bias distribution of novel classes with few-shot samples. By generating more samples from the calibrated distribution, EDC can provide sufficient supervision for subsequent classifier training. The proposed method is evaluated on three public skin disease datasets(i.e., ISIC2018, Derm7pt, and SD198), achieving significant performance improvements over state-of-the-art methods.

SLTRN: Sample-level transformer-based relation network for few-shot classification

Few-shot classification recognizes novel categories with limited labeled samples. The classic Relation Network (RN) compares support-query sample pairs for few-shot classification but overlooks support set contextual information, limiting its comparison capabilities. This work reformulates learning the relationship between query samples and each support class as a seq2seq problem. We introduce a Sample-level Transformer-based Relation Network (SLTRN) that utilizes sample-level self-attention to enhance the comparison ability of the relationship module by mining potential relationships among support classes. SLTRN demonstrates comparable performance with state-of-the-art methods on benchmarks, particularly excelling in the 1-shot setting with 52.11% and 67.55% accuracy on miniImageNet and CUB, respectively. Extensive ablation experiments validate the effectiveness and optimal settings of SLTRN. The experimental code for this work is available at https://github.com/ZitZhengWang/SLTRN.

A combination network based on graph modules for cardiac segmentation

BackgroundCardiac segmentation plays a crucial role in extracting cardiac substructures from computed tomography images, providing necessary information for subsequent cardiac disease diagnostics. Current research on cardiac segmentation algorithms faces challenges such as complex spatial distributions and limited labeled samples. The preferred method, convolutional neural network (CNN), excels at capturing features at various levels. However, the coarse convolutional operations of CNN often neglect valuable structural information. In contrast, graph neural network (GNN) excels at capturing complex spatial relationships among nodes by aggregating messages from graphs. MethodologyIn this work, a GNN-CNN combination network (GCCN) with enhanced feature extraction by GNN is proposed for cardiac segmentation. The combination network integrates the strengths of two approaches while compensating for each other's shortcomings. Firstly, 3D CT images are transformed into graphs, whose nodes with similar feature values within a certain spatial proximity are connected. Secondly, a lightweight graph attention network (LGAT) module is proposed for aggregating the messages of constructed graphs. Thirdly, the data, resampled from the feature-enhanced graphs, is fed into the 3D ResU-Net for the final segmentation process. ResultThe proposed method achieves an average Dice score of 92.98% which is 1.84% higher than 3D ResU-Net. The average Dice score of each cardiac substructure improved, particularly for the ascending aorta, achieving 96.35%, an 8.29% improvement over 3D ResU-Net. Additionally, our study demonstrates that GCCN has the improved adaptability in handling datasets with limited samples. ConclusionIn general, GCCN allows for the extraction of richer structural information from CT images, resulting in stronger feature representation of the data. Compared to pure convolutional methods, GCCN not only enhances segmentation accuracy but also exhibits greater adaptability to small-sample data. GCCN represents an exploration of applying graph neural networks to cardiac segmentation tasks from a novel perspective.

Boosting efficient attention assisted cyclic adversarial auto-encoder for rotating component fault diagnosis under low label rates

In practical engineering scenarios with limited labeled samples, conventional semi-supervised diagnostic methods face challenges in achieving satisfactory identification outcomes. To address the aforementioned issues, this paper introduces an incremental semi-supervised learning (ISL) approach based on boosting efficient attention (BEA) assisted cyclic adversarial auto-encoder (CAAE), referred to as BEA-CAAE. The CAAE enhances unsupervised feature representation by simultaneously constraining the distribution of encoded features and aligning elements of the reconstructed samples through a cyclic encoding strategy. The BEA improves classical attention weights' activation strength to better capture vital information, thereby boosting the feature extraction capabilities of both CAAE and the classifier. The ISL employs a stepwise pseudo-label propagation strategy to incrementally filter high-confidence samples, enhancing sample and label utilization, and improving diagnostic accuracy under low label rate conditions. Experiments conducted on multiple test rigs with simple structures as well as large-scale rotating components test rigs that mimic real-world engineering conditions have demonstrated that the proposed method exhibits a significant advantage over existing semi-supervised fault diagnosis approaches in terms of fault diagnosis accuracy and generalization capability, especially under low label rate conditions.

SSPENet: Semi-supervised prototype enhancement network for rolling bearing fault diagnosis under limited labeled samples

Real industrial scenarios struggle with the issues of a limited number of labeled samples and difficulty in accessing, which results in deep learning-based fault diagnosis models having poor generalization capabilities and decreased diagnostic accuracy. To address this problem, a semi-supervised prototype enhancement network (SSPENet) is proposed for rolling bearing fault diagnosis in this study. Firstly, a dual pooling attention residual network is proposed to be used in the feature extraction module. The goal is to efficiently extract the hidden features within rolling bearings, thus enabling the accurate classification of different sample categories. Subsequently, the Hungarian algorithm is utilized to design a strategic approach to update prototypes with pseudo-labels, which achieves the effect of augmenting prototypes by accurately adjusting the prototype position of each class of limited labeled samples through unlabeled samples, to improve the discriminative ability of the network model for fault classes. Finally, validation and experimental analysis are carried out on two bearing datasets, which achieve the average diagnostic accuracy of the proposed model to be above 90 % for both 1-shot and 2-shot cases, obtaining more satisfactory diagnostic results.

A limited labeled data bearing fault diagnosis method based on self-supervised learning

The key focus of this study revolves around several crucial issues concerning bearing fault diagnosis, and presents an innovative solution. Conventional labeled approaches for bearing fault diagnosis often necessitate labeled data sets, which can be time-consuming or infeasible to obtain. To address this problem, an increasing amount of research has started to explore fault diagnosis methods that utilize limited labeled data. In our study, we introduce a framework for bearing fault diagnosis that incorporates wavelet transform and self-supervised learning techniques. The framework leverages vibration signals and transforms them into time-frequency spectrograms as inputs. To extract features, we employ the Swin Transformer as an encoder. Furthermore, we present a self-supervised learning approach named MoBy to address the challenge of limited labeled samples. Encouragingly, our approach achieves a diagnostic accuracy of 96.4% by utilizing only 1% labeled samples, through a well-trained encoder and a simple linear classification layer. This demonstrates outstanding performance in utilizing limited labeled data. To validate the superiority of our proposed approach, we conducted experiments on two rolling bearing fault datasets and achieved significant results.

Liberating Seen Classes: Boosting Few-Shot and Zero-Shot Text Classification via Anchor Generation and Classification Reframing

Few-shot and zero-shot text classification aim to recognize samples from novel classes with limited labeled samples or no labeled samples at all. While prevailing methods have shown promising performance via transferring knowledge from seen classes to unseen classes, they are still limited by (1) Inherent dissimilarities among classes make the transformation of features learned from seen classes to unseen classes both difficult and inefficient. (2) Rare labeled novel samples usually cannot provide enough supervision signals to enable the model to adjust from the source distribution to the target distribution, especially for complicated scenarios. To alleviate the above issues, we propose a simple and effective strategy for few-shot and zero-shot text classification. We aim to liberate the model from the confines of seen classes, thereby enabling it to predict unseen categories without the necessity of training on seen classes. Specifically, for mining more related unseen category knowledge, we utilize a large pre-trained language model to generate pseudo novel samples, and select the most representative ones as category anchors. After that, we convert the multi-class classification task into a binary classification task and use the similarities of query-anchor pairs for prediction to fully leverage the limited supervision signals. Extensive experiments on six widely used public datasets show that our proposed method can outperform other strong baselines significantly in few-shot and zero-shot tasks, even without using any seen class samples.

A hybrid convolution transformer for hyperspectral image classification

ABSTRACT Hyperspectral images play a crucial role in remote sensing applications surveillance, environment and precision agriculture, containing abundant object information. However, they often face challenges such as limited labelled data and imbalanced classes. In recent years, convolutional neural networks (CNNs) have shown impressive performance in computer vision tasks, including hyperspectral image classification. The emergence of transformers has also attracted attention for hyperspectral image analysis due to their promising capabilities. Nevertheless, transformers typically demand a substantial amount of training data, making their application challenging in scenarios with limited labelled samples. To overcome this limitation, we propose a hybrid convolution transformer framework. Our method uses a vision transformer and a residual 3D convolutional neural network model. It also uses a sequence aggregation layer to avoid overfitting issues that come up when there isn’t enough training data. Our proposed residual channel attention module captures richer spatial-spectral complementary information and maintains spectral details during the feature extraction process. We conducted experiments on three benchmark datasets. The proposed model achieved state of the art performance 99.75 % , 99.46 % and 99.95 % in terms of overall accuracy (OA) using only 5 % , 5 % and 10 % labelled training samples respectively. Which is better than other state of the art methods.

TR-RAGCN-AFF-RESS: A Method for Radar Emitter Signal Sorting

Radar emitter signal sorting (RESS) is a crucial process in contemporary electronic battlefield situation awareness. Separating pulses belonging to the same radar emitter from interleaved radar pulse sequences with a lack of prior information, high density, strong overlap, and wide parameter distribution has attracted increasing attention. In order to improve the accuracy of RESS under scenarios with limited labeled samples, this paper proposes an RESS model called TR-RAGCN-AFF-RESS. This model transforms the RESS problem into a pulse-by-pulse classification task. Firstly, a novel weighted adjacency matrix construction method was proposed to characterize the structural relationships between pulse attribute parameters more accurately. Building upon this foundation, two networks were developed: a Transformer(TR)-based interleaved pulse sequence temporal feature extraction network and a residual attention graph convolutional network (RAGCN) for extracting the structural relationship features of attribute parameters. Finally, the attention feature fusion (AFF) algorithm was introduced to fully integrate the temporal features and attribute parameter structure relationship features, enhancing the richness of feature representation for the original pulses and achieving more accurate sorting results. Compared to existing deep learning-based RESS algorithms, this method does not require many labeled samples for training, making it better suited for scenarios with limited labeled sample availability. Experimental results and analysis confirm that even with only 10% of the training samples, this method achieves a sorting accuracy exceeding 93.91%, demonstrating high robustness against measurement errors, lost pulses, and spurious pulses in non-ideal conditions.

LinkNet-Spectral-Spatial-Temporal Transformer Based on Few-Shot Learning for Mangrove Loss Detection with Small Dataset

Mangroves grow in intertidal zones in tropical and subtropical regions, offering numerous advantages to humans and ecosystems. Mangrove monitoring is one of the important tasks to understand the current status of mangrove forests regarding their loss issues, including deforestation and degradation. Currently, satellite imagery is widely employed to monitor mangrove ecosystems. Sentinel-2 is an optical satellite imagery whose data are available for free, and which provides satellite imagery at a 5-day temporal resolution. Analyzing satellite images before and after loss can enhance our ability to detect mangrove loss. This paper introduces a LSST-Former model that considers the situation before and after mangrove loss to categorize non-mangrove areas, intact mangroves, and mangrove loss categories using Sentinel-2 images for a limited number of labels. The LSST-Former model was developed by integrating a fully convolutional network (FCN) and a transformer base with few-shot learning algorithms to extract information from spectral-spatial-temporal Sentinel-2 images. The attention mechanism in the transformer algorithm may effectively mitigate the issue of limited labeled samples and enhance the accuracy of learning correlations between samples, resulting in more successful classification. The experimental findings demonstrate that the LSST-Former model achieves an overall accuracy of 99.59% and an Intersection-over-Union (IoU) score of 98.84% for detecting mangrove loss, and the validation of universal applicability achieves an overall accuracy of more than 92% and a kappa accuracy of more than 89%. LSST-Former demonstrates superior performance compared to state-of-the-art deep-learning models such as random forest, Support Vector Machine, U-Net, LinkNet, Vision Transformer, SpectralFormer, MDPrePost-Net, and SST-Former, as evidenced by the experimental results and accuracy metrics.

S3L: Spectrum Transformer for Self-Supervised Learning in Hyperspectral Image Classification

In the realm of Earth observation and remote sensing data analysis, the advancement of hyperspectral imaging (HSI) classification technology is of paramount importance. Nevertheless, the intricate nature of hyperspectral data, coupled with the scarcity of labeled data, presents significant challenges in this domain. To mitigate these issues, we introduce a self-supervised learning algorithm predicated on a spectral transformer for HSI classification under conditions of limited labeled data, with the objective of enhancing the efficacy of HSI classification. The S3L algorithm operates in two distinct phases: pretraining and fine-tuning. During the pretraining phase, the algorithm learns the spatial representation of HSI from unlabeled data, utilizing a masking mechanism and a spectral transformer, thereby augmenting the sequence dependence of spectral features. Subsequently, in the fine-tuning phase, labeled data is employed to refine the pretrained weights, thereby improving the precision of HSI classification. Within the comprehensive encoder–decoder framework, we propose a novel spectral transformer module specifically engineered to synergize spatial feature extraction with spectral domain analysis. This innovative module adeptly navigates the complex interplay among various spectral bands, capturing both global and sequential spectral dependencies. Uniquely, it incorporates a gated recurrent unit (GRU) layer within the encoder to enhance its ability to process spectral sequences. Our experimental evaluations across several public datasets reveal that our proposed method, distinguished by its spectral transformer, achieves superior classification performance, particularly in scenarios with limited labeled samples, outperforming existing state-of-the-art approaches.