Sample Imbalance Problem Research Articles

Document-Level Relation Extraction with Uncertainty Pseudo-Label Selection and Hard-Sample Focal Loss

Relation extraction is a fundamental task in natural language processing that aims to identify structured triple relationships from unstructured text. In recent years, research on relation extraction has gradually advanced from the sentence level to the document level. Most existing document-level relation extraction (DocRE) models are fully supervised and their performance is limited by the dataset quality. However, existing DocRE datasets suffer from annotation omission, making fully supervised models unsuitable for real-world scenarios. To address this issue, we propose the DocRE method based on uncertainty pseudo-label selection. This method first trains a teacher model to annotate pseudo-labels for a dataset with incomplete annotations, trains a student model on the dataset with annotated pseudo-labels, and uses the trained student model to predict relations on the test set. To mitigate the confirmation bias problem in pseudo-label methods, we performed adversarial training on the teacher model and calculated the uncertainty of the model output to supervise the generation of pseudo-labels. In addition, to address the hard-easy sample imbalance problem, we propose an adaptive hard-sample focal loss. This loss can guide the model to reduce attention to easy-to-classify samples and outliers and to pay more attention to hard-to-classify samples. We conducted experiments on two public datasets, and the results proved the effectiveness of our method.

I5mC-DCGA: an improved hybrid network framework based on the CBAM attention mechanism for identifying promoter 5mC sites

Background5-Methylcytosine (5mC) plays a very important role in gene stability, transcription, and development. Therefore, accurate identification of the 5mC site is of key importance in genetic and pathological studies. However, traditional experimental methods for identifying 5mC sites are time-consuming and costly, so there is an urgent need to develop computational methods to automatically detect and identify these 5mC sites.ResultsDeep learning methods have shown great potential in the field of 5mC sites, so we developed a deep learning combinatorial model called i5mC-DCGA. The model innovatively uses the Convolutional Block Attention Module (CBAM) to improve the Dense Convolutional Network (DenseNet), which is improved to extract advanced local feature information. Subsequently, we combined a Bidirectional Gated Recurrent Unit (BiGRU) and a Self-Attention mechanism to extract global feature information. Our model can learn feature representations of abstract and complex from simple sequence coding, while having the ability to solve the sample imbalance problem in benchmark datasets. The experimental results show that the i5mC-DCGA model achieves 97.02%, 96.52%, 96.58% and 85.58% in sensitivity (Sn), specificity (Sp), accuracy (Acc) and matthews correlation coefficient (MCC), respectively.ConclusionsThe i5mC-DCGA model outperforms other existing prediction tools in predicting 5mC sites, and it is currently the most representative promoter 5mC site prediction tool. The benchmark dataset and source code for the i5mC-DCGA model can be found in https://github.com/leirufeng/i5mC-DCGA.

Using data resampling and category weight adjustment to solve sample imbalance

The purpose of this thesis is to investigate the application of artificial intelligence and machine learning in solving the sample imbalance problem. The sample imbalance problem refers to the phenomenon that the number of different categories of samples in the training data varies greatly, resulting in the poor performance of traditional machine learning algorithms on a few categories of samples. To address this problem, this paper proposes a new approach combining data resampling and category weight adjustment strategies. First, the sample distribution of the dataset is adjusted by undersampling and oversampling techniques to balance the number of samples from different categories. Then, during the model training process, different weights are assigned to the samples of different categories so that the model pays more attention to the samples of a few categories. The experimental results show that the method achieves significant performance improvement on multiple datasets. In addition, this paper compares other commonly used methods for solving the sample imbalance problem and analyzes and discusses them in detail. Finally, this study offers a practical solution to the problem of sample imbalance and provides guidance for research in related fields.

CycleGAN-Based Data Augmentation for Subgrade Disease Detection in GPR Images with YOLOv5

Vehicle-mounted ground-penetrating radar (GPR) technology is an effective means of detecting railway subgrade diseases. However, existing methods of GPR data interpretation largely rely on manual identification, which is not only inefficient but also highly subjective. This paper proposes a semi-supervised deep learning method to identify railway subgrade diseases. This method addresses the sample imbalance problem in the defect dataset by utilizing a data augmentation method based on a generative adversarial network model. An initial network model for disease identification is obtained by training the YOLOv5 network with a small number of existing samples. The intelligently extended samples are then labeled to achieve a balance in the disease samples. The network is trained to improve the recognition accuracy of the intelligent model using a more complete dataset. The experimental results show that the accuracy of the proposed method can reach up to 94.53%, which is 23.85% higher than that of the supervised learning model without an extended dataset. This has strong industrial application value for railway subgrade disease detection as the potential learning ability of the model can be explored to a greater extent, thereby improving the recognition accuracy of subgrade diseases.

Application of LightGBM hybrid model based on TPE algorithm optimization in sleep apnea detection.

Sleep apnoea syndrome (SAS) is a serious sleep disorder and early detection of sleep apnoea not only reduces treatment costs but also saves lives. Conventional polysomnography (PSG) is widely regarded as the gold standard diagnostic tool for sleep apnoea. However, this method is expensive, time-consuming and inherently disruptive to sleep. Recent studies have pointed out that ECG analysis is a simple and effective diagnostic method for sleep apnea, which can effectively provide physicians with an aid to diagnosis and reduce patients' suffering. To this end, in this paper proposes a LightGBM hybrid model based on ECG signals for efficient detection of sleep apnea. Firstly, the improved Isolated Forest algorithm is introduced to remove abnormal data and solve the data sample imbalance problem. Secondly, the parameters of LightGBM algorithm are optimised by the improved TPE (Tree-structured Parzen Estimator) algorithm to determine the best parameter configuration of the model. Finally, the fusion model TPE_OptGBM is used to detect sleep apnoea. In the experimental phase, we validated the model based on the sleep apnoea ECG database provided by Phillips-University of Marburg, Germany. The experimental results show that the model proposed in this paper achieves an accuracy of 95.08%, a precision of 94.80%, a recall of 97.51%, and an F1 value of 96.14%. All of these evaluation indicators are better than the current mainstream models, which is expected to assist the doctor's diagnostic process and provide a better medical experience for patients.

A stacking-based ensemble learning model for kneading paste quality intelligent prediction: A real case study

Paste kneading is a vital process of prebaked carbon anode production, and the quality of the paste has a great impact on the quality of the final product. However, it is difficult to inspect the paste quality in line. Because the inspection of the paste quality in the laboratory is not real-time, the paste has already entered the next process after the results are obtained. And the manual quality inspection is labor-intensive and unsafe. Therefore, a stacking-based ensemble learning model for kneading paste quality prediction is proposed in this paper. The gradient boosting decision tree, random forest, k-nearest neighbors, and support vector machine are used as base learners, and logistic regression is used as meta-learner. Kneading production data of 572 paste pots are collected for quality prediction, where each pot of paste data contains 44 signals. The correlation coefficient-based feature engineering method was applied, and 10 features with the greatest correlation with paste quality were identified to construct the dataset. The up-sampling and under-sampling methods are used to solve the problem of sample imbalance. Parallel comparison is applied to verify the advantage of the stacking-based ensemble learning model, and the results indicate that the model performs better than every single classifier and has higher accuracy and generalization ability, especially for imbalanced samples.

Optical gas imaging for leak detection based on improved deeplabv3+ model

Optical gas imaging (OGI) is a technique to dynamically observe gas leaks with a wide detection range. As gasses are mobile and diffusive, traditional motion detection methods are difficult to adapt to the change of gas motion state with the environment, resulting in poor segmentation effect and low accuracy. Therefore, this paper proposes a deep learning-based semantic segmentation method of gas for optical gas imaging detection based on DeeplabV3+ model. Firstly, MobileNetV2 is used to replace the original backbone network to reduced model parameters and complexity and facilitates the edge deployment of the model. Second, the dense atrous spatial pyramid pooling (DenseASPP) module is introduced to improve the atrous spatial pyramid pooling (ASPP) module of the model to increase the model sensing field and improve the model gas segmentation accuracy. Finally, Dice loss is used to solve the sample imbalance problem. The experimental results show that IOU and F1 score of the model are improved by 4.10% and 2.47%, and it provides a new intelligent detection method for optical gas imaging.

Electrical insulator defect detection with incomplete annotations and imbalanced samples

AbstractInsulators are one of the key components in high‐voltage power systems that prevent transmission lines from grounding. Since they are exposed to different kinds of harsh environments and climates, periodic inspection is indispensable for the safety and high quality of power grid. Nowadays, unmanned aerial vehicle (UAV) inspection is more widely used, facilitating incorporation of convolutional neural network‐based detectors in the insulator detection task. However, these methods are generally based on the assumption that the image samples are balanced among different categories and possess completely ideal annotations. The problem of sample imbalance or incomplete annotation is rarely investigated in depth for insulator defect detection. Here, insulator defect detection with imbalanced data and incomplete annotations is focused on. The proposed framework, named Pi‐index, introduces positive unlabelled (PU) learning to solve the problem of incomplete annotation and designs a novel index the class prior, which is a key parameter in PU learning. Moreover, focal loss is integrated in our framework to alleviate the effect of sample imbalance. Experiment results demonstrate that the proposed framework achieves better performance than the baseline methods in situations of sample imbalance and missing annotation.

Research of data mining methods for classification of imbalanced data sets

With the rapid development of information technology, which is widely used in all spheres of human life and activity, extremely large amounts of data have been accumulated today. By applying machine learning methods to this data, new practically useful knowledge can be obtained. The main goal of this paper is to study different machine learning methods for solving the classification problem and compare their efficiency and accuracy. A separate task is data pre-processing aimed at solving the problem of sample imbalance, as well as identifying the principal components that will be used to solve the classification problem. For this purpose, an information system for classifying the bankruptcy of a company with specified economic and financial characteristics was researched and developed. The study uses a dataset on the basis of which the efficiency and quality of application of several existing classification algorithms are evaluated. These classifiers are: conventional and linear Support Vector Machine, Extra Trees, Random Forest, Decision Tree, Logistic Regression, Multilayer perceptron Classifier, Gradient Boosting, Naive Bayes Classifier. For data pre-processing, we scaled the data, used the SMOTE method to get rid of the imbalance of the training sample, and performed principal component analysis and L1 regularisation. Principal component analysis allowed us to identify 15 principal components that have the greatest impact on classification accuracy and, accordingly, use them in the classification process. Analysing the results, we found that the best classifier was Random Forest with 95.9 % accuracy, and the worst was Naive Bayes with 85.1 %. To evaluate the quality of classification and select the best classifier, the Confusion matrix is used, which takes into account the number of true positive (TP) and true negative (TN) values, as well as the number of false negative (FN) and false positive (FP) classification results, and the values of such metrics as accuracy, precision, sensitivity, F1, and ROC. Accuracy is the percentage of correct answers given by the algorithm, while Recall is the number of TPs divided by the number of TPs plus the number of FNs. F1 indicates the balance between accuracy and sensitivity. Precision is the number of true positive predictions divided by the number of false positive and true negative predictions. ROC AUC is a tool for measuring performance for classification tasks at different thresholds. It shows how well a model can distinguish between classes. The conclusions present the main results of the study and indicate the main future direction of the work, namely, the study of classification results for other datasets and more efficient processing and analysis.

Apple Leaf Disease Detection Based on Lightweight YOLOv8-GSSW

HighlightsModify the YOLOV8 backbone network to reduce model calculation complexity and parameter amount.Integrating the Slim-Neck design and the SA attention mechanism into the neck network to enhance the attention of target features and improve the weight of the target.Adoption of the WIoU loss function over the original CIoU to address the sample imbalance problem and accelerate model convergence.Abstract. In response to the challenges posed by the excessive number of parameters and the high computational complexity inherent in existing algorithms for apple leaf disease recognition, this study introduces an enhanced, lightweight detection algorithm, YOLOv8-GSSW. Building upon the YOLOv8n as the foundational model, this research innovates in several key areas. Initially, the algorithm employs a lightweight GSConv convolutional module to replace the standard convolutional modules within the backbone network, significantly reducing both the model’s computational complexity and its parameter count. Furthermore, the incorporation of the Slim-Neck design concept into the neck network module serves to bolster the model’s capabilities in feature processing and learning. This enhancement is further augmented by the integration of the SA attention mechanism within Slim-Neck, which amplifies the attention to target features and increases their weight by merging information across both spatial and channel dimensions. Finally, the original CIoU loss function is substituted with the WIoU loss function to address the issue of sample imbalance and accelerate the model’s convergence. Experimental results underscore the YOLOv8-GSSW algorithm’s superior performance, achieving a mean average precision (mAP) of 87.3% with a parameter volume of 2.7×106 and a model weight of 5.4 MB. Relative to the YOLOv8n benchmark model, this represents a 15.6% reduction in parameter volume and an 11.5% decrease in model size. The modifications introduced in this algorithm not only facilitate more convenient application in practical scenarios for apple leaf disease detection but also render the algorithm well-suited for deployment on edge devices. Keywords: Apple leaves, Disease detection, Lightweight, YOLOv8n.

An end-to-end seed vigor prediction model for imbalanced samples using hyperspectral image.

Hyperspectral imaging is a key technology for non-destructive detection of seed vigor presently due to its capability to capture variations of optical properties in seeds. As the seed vigor data depends on the actual germination rate, it inevitably results in an imbalance between positive and negative samples. Additionally, hyperspectral image (HSI) suffers from feature redundancy and collinearity due to its inclusion of hundreds of wavelengths. It also creates a challenge to extract effective wavelength information in feature selection, however, which limits the ability of deep learning to extract features from HSI and accurately predict seed vigor. Accordingly, in this paper, we proposed a Focal-WAResNet network to predict seed vigor end-to-end, which improves the network performance and feature representation capability, and improves the accuracy of seed vigor prediction. Firstly, the focal loss function is utilized to adjust the loss weights of different sample categories to solve the problem of sample imbalance. Secondly, a WAResNet network is proposed to select characteristic wavelengths and predict seed vigor end-to-end, focusing on wavelengths with higher network weights, which enhance the ability of seed vigor prediction. To validate the effectiveness of this method, this study collected HSI of maize seeds for experimental verification, providing a reference for plant breeding. The experimental results demonstrate a significant improvement in classification performance compared to other state-of-the-art methods, with an accuracy up to 98.48% and an F1 score of 95.9%.

A three-stage pavement image crack detection framework with positive sample augmentation

Most pavement crack detection methods based on deep learning rely too much on pixel-wise labels, and are facing with sample imbalance problem. This paper proposes a three-stage pavement crack detection framework with positive sample augmentation by an autoencoder-deep convolutional generative adversarial network (E-DCGAN). Positive samples are the images labeled as crack in the training datasets. Firstly, crack regions are located using graph based on visual saliency (GBVS). Secondly, a convolutional neural network (CNN) model is designed to recognize crack pixels in the located crack regions. The E-DCGAN is explored to generate crack images, which are then added as augmented positive samples in the training dataset of the CNN model. The trained CNN model takes the located crack regions as inputs, outputs the recognized crack pixels. Finally, fine crack detection is realized through region growing. Crack contours are completely depicted pixel-wise based on the recognized crack pixels and the Breadth-first search. Experiments are carried out on two benchmark datasets and in practical applications. The results showed that the framework could detect cracks at the pixel level. Its performance is improved compared with standard fully convolutional network (FCN)-based and U-shaped CNN (U-Net)-based crack detection methods. In addition, the standard FCN-based and U-Net-based methods require accurate pixel-wise annotations, while the proposed framework only requires image-wise annotations. It can be concluded that the proposed framework not only effectively avoids the dependence on pixel-level annotations, improves the sample imbalance issues, but also achieves competitive detection performance over the FCN and U-Net based methods.

A novel network training approach for solving sample imbalance problem in Photovoltaic power prediction

Randomness and intermittency are crucial challenges in photovoltaic (PV) power prediction. Most studies concentrate on addressing the randomness of PV power, and tend to overlook the intermittency that leads to sample imbalance, which negatively affects prediction accuracy. To address the sample imbalance, a novel approach called segment imbalance regression (SIR) is proposed. The SIR method proactively exploits the inherent imbalanced nature of samples by investigating the interactions among neighbouring samples, which leads to dynamical assigning weights. Through focused training and segmental prediction, SIR selectively retains the outside information while focusing segment inside, which enhances the gradient descent process and ultimately leads to improved training performance. With crisscross optimization (CSO), SIR demonstrates its performance sufficiently with an average RMSE reduction of 21.17% and 40.76% in the multi-step prediction and day-ahead prediction cases, respectively.

An automated defect detection method for optimizing industrial quality inspection

Detecting defects in industrial-quality inspection is an important task, but defect detection remains challenging due to limited size, various types of defects, and imbalanced samples in images. In this paper, we propose a novel automatic defect detection network that efficiently detects defects. We introduce the Automated Sample Assignment (SCID) algorithm, which automatically divides the distribution of positive and negative samples based on similarity scores using the Gaussian mixture model. The Automated Sample Assignment (SCID) algorithm solves the problem of sample imbalance in defect detection networks caused by manually setting the intersection over the union (IOU) threshold and assigning positive and negative labels to samples. Then, instead of the pooling method used by most backbone networks, we use a loss fusion pool named LE-Pool to retain more detailed information. Finally, we use the soft weighted attention mechanism of the detection head (S-head) to identify noise information. Experimental results on multiple defect detection datasets show that the proposed method is superior to other detection methods in terms of accuracy.

YOLO-Crater Model for Small Crater Detection

Craters are the most prominent geomorphological features on the surface of celestial bodies, which plays a crucial role in studying the formation and evolution of celestial bodies as well as in landing and planning for surface exploration. Currently, the main automatic crater detection models and datasets focus on the detection of large and medium craters. In this paper, we created 23 small lunar crater datasets for model training based on the Chang’E-2 (CE-2) DOM, DEM, Slope, and integrated data with 7 kinds of visualization stretching methods. Then, we proposed the YOLO-Crater model for Lunar and Martian small crater detection by replacing EioU and VariFocal loss to solve the crater sample imbalance problem and introducing a CBAM attention mechanism to mitigate interference from the complex extraterrestrial environment. The results show that the accuracy (P = 87.86%, R = 66.04%, and F1 = 75.41%) of the Lunar YOLO-Crater model based on the DOM-MMS (Maximum-Minimum Stretching) dataset is the highest and better than that of the YOLOX model. The Martian YOLO-Crater, trained by the Martian dataset from the 2022 GeoAI Martian Challenge, achieves good performance with P = 88.37%, R = 69.25%, and F1 = 77.65%. It indicates that the YOLO-Crater model has strong transferability and generalization capability, which can be applied to detect small craters on the Moon and other celestial bodies.

Off-Road Environment Semantic Segmentation for Autonomous Vehicles Based on Multi-Scale Feature Fusion

For autonomous vehicles driving in off-road environments, it is crucial to have a sensitive environmental perception ability. However, semantic segmentation in complex scenes remains a challenging task. Most current methods for off-road environments often have the problems of single scene and low accuracy. Therefore, this paper proposes a semantic segmentation network based on LiDAR called Multi-scale Augmentation Point-Cylinder Network (MAPC-Net). The network uses a multi-layer receptive field fusion module to extract features from objects of different scales in off-road environments. Gated feature fusion is used to fuse PointTensor and Cylinder for encoding and decoding. In addition, we use CARLA to build off-road environments for obtaining datasets, and employ linear interpolation to enhance the training data to solve the problem of sample imbalance. Finally, we design experiments to verify the excellent semantic segmentation ability of MAPC-Net in an off-road environment. We also demonstrate the effectiveness of the multi-layer receptive field fusion module and data augmentation.

Malfunction diagnosis of main station of power metering system using LSTM-ResNet with SMOTE method

The power metering system is an important part of the smart grid for data acquisition and analysis. The fault state of the main station directly affects the stable and safe operation of the power metering system. Hinged on the real-world data supplied by the monitoring platform of the Metrology Center of Guangdong Power Grid Co., Ltd., we present a novel malfunction diagnosis method for the main station of the power metering system. The proposed method utilizes the synthetic mi-nority over-sampling technique (SMOTE) and designs a combined model of long short-term memory (LSTM) network and ResNet. SMOTE solves the sample imbalance problem. Furthermore, the combined LSTM-ResNet model employs LSTM to extract the time-dependent signal feature and exploits ResNet to optimize data flow. Consequently, the proposed LSTM-ResNet model improves training efficiency and malfunction diagnosis accuracy. The proposed diagnosis mthod is verifird on the real-world data, which proves the proposed method’s surpass traditional methods. A specific analysis of results and the practical application of the proposed method is also elaborated.

Inter-patient ECG heartbeat classification for arrhythmia classification: a new approach of multi-layer perceptron with weight capsule and sequence-to-sequence combination.

Objective: The objective of this research is to construct a method to alleviate the problem of sample imbalance in classification, especially for arrhythmia classification. This approach can improve the performance of the model without using data enhancement. Methods: In this study, we have developed a new Multi-layer Perceptron (MLP) block and have used a Weight Capsule (WCapsule) network with MLP combined with sequence-to-sequence (Seq2Seq) network to classify arrhythmias. Our work is based on the MIT-BIH arrhythmia database, the original electrocardiogram (ECG) data is classified according to the criteria recommended by the American Association for Medical Instrumentation (AAMI). Also, our method's performance is further evaluated. Results: The proposed model is evaluated using the inter-patient paradigm. Our proposed method shows an accuracy (ACC) of 99.88% under sample imbalance. For Class N, sensitivity (SEN) is 99.79%, positive predictive value (PPV) is 99.90%, and specificity (SPEC) is 99.19%. For Class S, SEN is 97.66%, PPV is 96.14%, and SPEC is 99.85%. For Class V, SEN is 99.97%, PPV is 99.07%, and SPEC is 99.94%. For Class F, SEN is 97.94%, PPV is 98.70%, and SPEC is 99.99%. When using only half of the training sample, our method shows that the SEN of Class N and V is 0.97% and 5.27% higher than the traditional machine learning algorithm. Conclusion: The proposed method combines MLP, weight capsule network with Seq2seq network, effectively addresses the problem of sample imbalance in arrhythmia classification, and produces good performance. Our method also shows promising potential in less samples.

Substation rotational object detection based on multi-scale feature fusion and refinement

In modern energy systems, substations are the core of electricity transmission and distribution. However, similar appearance and small size pose significant challenges for automatic identification of electrical devices. To address these issues, we collect and annotate the substation rotated device dataset (SRDD). Further, feature fusion and feature refinement network (F3RNet) are constructed based on the classic structure pattern of backbone-neck-head. Considering the similar appearance of electrical devices, the deconvolution fusion module (DFM) is designed to enhance the expression of feature information. The balanced feature pyramid (BFP) is embedded to aggregate the global feature. The feature refinement is constructed to adjust the original feature maps by considering the feature alignment between the anchors and devices. It can generate more accurate feature vectors. To address the problem of sample imbalance between electrical devices, the gradient harmonized mechanism (GHM) loss is utilized to adjust the weight of each sample. The ablation experiments are conducted on the SRDD dataset. F3RNet achieves the best detection performance compared with classical object detection networks. Also, it is verified that the features from global feature maps can effectively recognize the similar and small devices.

Active learning based on similarity level histogram and adaptive-scale sampling for very high resolution image classification

In remote sensing image classification, active learning aims to obtain an excellent classification model by selecting informative or representative training samples. However, due to the complexity of remote sensing images, the same class of ground objects usually have different spectral representations. The existing active learning methods may not take into account diverse representations of the same targets, which leads to a possible lack of intra-class diversity in the collected samples. To alleviate this problem, we propose an active learning method based on similarity level histogram (SLH) and adaptive-scale sampling to improve very high resolution remote sensing image classification. Specifically, we construct a SLH for each class of ground objects to effectively consider the intra-class diversity of the same target. To avoid the problem of sample imbalance caused by over-sampling or under-sampling, we design an adaptive-scale sampling strategy. Then, we utilize active learning to mine representative samples from each SLH warehouse according to adaptive-scale sampling strategies until the iteration condition is satisfied. Experiments show that the proposed algorithm can achieve better classification performance with limited training samples and is competitive with other methods based on four sets of publicly available data.