Multi-scale Convolutional Neural Network Research Articles

Knowledge graph embedding in a uniform space

Knowledge graph embedding (KGE) is typically used for link prediction to automatically predict missing links in knowledge graphs. Current KGE models are mainly based on complicated mathematical associations, which are highly expressive but ignore the uniformity behind the classical bilinear translational model TransE, a model that embeds all entities of knowledge graphs in a uniform space, enabling accurate embeddings. This study analyses the uniformity of TransE and proposes a novel KGE model called ConvUs that follows uniformity with expressiveness. Based on the convolution neural network (CNN), ConvUs proposes constraints on convolution filter values and employs a multi-layer, multi-scale CNN architecture with a non-parametric L2 norm-based scoring function for the calculation of triple scores. This addresses potential uniformity-related issues in existing CNN-based KGE models, allowing ConvUs to maintain a uniform embedding space while benefiting from the powerful expressiveness of CNNs. Furthermore, circular convolution is applied to alleviate the potential orderliness contradictions, making ConvUs more suitable for conducting uniform space KGE. Our model outperformed the base model ConvKB and several baselines on the link prediction benchmark WN18RR and FB15k-237, demonstrating strong applicability and generalization and indicating that the uniformity of embedding space with high expressiveness enables more efficient knowledge graph embeddings.

Series Arc Fault Detection under Vibration Condition Based on N-M-M-B

Under the conditions of a mechanical fault in a motor, mechanical vibration of a specific frequency can be generated. The electrical contact points directly connected to the motor can vibrate at the same frequency. The electrical contact points with poor contact can easily produce a series arc fault under vibration conditions, which affects the reliability of the power supply. In order to detect the series arc fault under different vibration conditions, the arc fault generator is connected between the back end of the frequency converter and the motor. An arc fault experiment under different vibration conditions was carried out and the fault phase current and arc voltage signals were collected. In this paper, the noise-assisted multivariate empirical mode decomposition and the correlation coefficient between each intrinsic mode function are used to select the fault feature signals. Then, the reconstructed signal is input into the series arc fault model combining a multi-scale convolutional neural network and a bidirectional long short-term memory network for training. The research results show that the series arc fault detection method proposed in this paper can effectively detect the series arc fault and can preliminarily identify the type of motor fault causing the mechanical vibration of the motor; the model has good noise immunity and generalization.

A proposed video super-resolution reconstruction strategy using wavelet multi-scale convolutional neural networks

A proposed video super-resolution reconstruction strategy using wavelet multi-scale convolutional neural networks

Accurate authentication based on ECG using deep learning

Biometric-based authentication methods have been widely used, for example on portable devices (e.g., Android and iOS devices). However, there are several known limitations in existing authentication methods based on biometrics (e.g., those using facial, iris, and fingerprint). For example, in a healthcare context, a user may be physically incapable of completing the authentication due to his/her medical conditions. Hence, as a complementary authentication mechanism, there have been attempts to also utilize electrocardiogram (ECG). In this work, we propose an ECG authentication system that leverages deep learning. Specifically, to achieve generalization ability, complementary ensemble empirical decomposition (CEEMD) is introduced in our design. Moreover, a 1-D Multi-scale Convolutional Neural Network (1-D MCNN) is implemented to achieve accurate authentication. To evaluate the usability of our proposed approach, we have performed extensive experiments on eight databases, and the findings show that our proposed approach achieves good performance even on abnormal databases and can be adapted for different application environments. In addition, our adopted data from eight public databases requires theoretical statistical treatment for practical applications in real authentication scenarios.

Near-infrared quality monitoring modeling with multi-scale CNN and temperature adaptive correction

Considering the impact of temperature on near-infrared spectroscopy (NIRS), a temperature adaptive correction neural network model called TI-CNN has been developed. This model takes into account the temperature influence. First, a multi-scale CNN model is established to extract the global features of NIRS. Then, by establishing the quantitative relationship between the sample and the modeling error of the neural network, the influence of temperature on the spectral features. Based on the degree of influence, the weight is adaptively adjusted to quantify the effect of temperature on near-infrared spectral features. Finally, to verify the model performance, EPO-PLSR, OSC-PLSR, CNN, and TA-CNN were used for comparison on the test set. The results showed that traditional temperature correction models performed poorly, with a coefficient of determination R2< 0.80. However, the neural network models based on multi-scale CNN, R2> 0.80. Of these, the TI-CNN based on multi-scale CNN and temperature adaptive correction, demonstrated the best performance with R2> 0.95. Therefore, the method proposed in this paper demonstrates superior performance compared to existing methods. It utilizes a multi-scale CNN to extract near-infrared spectral features and also incorporates temperature to adaptively correct the spectral features, resulting in superior model performance across various temperature conditions.

TECD_Attention: Texture-enhanced and cross-domain attention modeling for visual place recognition

Visual place recognition (VPR) is a challenging task for visual computing in the field of robot navigation. However, most of the existing methods fail to learn the most salient features of place images by simple CNN feature or popular Transformer feature due to the inconsistency problem commonly existing in VPR datasets, which limits the robustness and interpretability of the model. In addition, existing state-of-the-art methods only capture general features of original places with multi-scale CNN or transformer features and ignore texture characteristics existing in place images, resulting in suboptimal recognition performance. To cope with the above issues, we propose a novel visual place recognition network, named Texture-enhanced Cross-domain Attention Transformer (TECD_Attention). Specially, a cross-attention Transformer is first used for fusing deep attentive local and global features to improve the multi-scale feature representation of the recognition model. Second, a texture-enhanced cross-domain attention block is designed to construct the final feature descriptor by fusing texture features and attentive local–global features. Then, a tripled loss function is used for matching top-ranked reference places from the place database to a query place. Last, effective and efficient place re-ranking is achieved by training an adapted weakly supervised re-ranking network relying on the similarity computing between the query place and the top-ranked places. Our approach is carried out in extensive experiments on four challenging datasets. Our model has achieved 96.2%, 94.6%, 95.9%, and 96.8% average recall based on top 1% Candidate scenario on Tokyo 24/7, Pitts250k, VPRiCE, and SUN397 datasets, respectively. Therefore, Compared with the existing state-of-the-art VPR methods, TECD_Attention performs superior on robot place recognition in challenging environments. Hence, we can conclude that this is a robust model for robot visual place recognition in challenging environments.

An Electrocardiogram Classification Using a Multiscale Convolutional Causal Attention Network

Electrocardiograms (ECGs) play a pivotal role in the diagnosis and prediction of cardiovascular diseases (CVDs). However, traditional methods for ECG classification involve intricate signal processing steps, leading to high design costs. Addressing this concern, this study introduces the Multiscale Convolutional Causal Attention network (MSCANet), which utilizes a multiscale convolutional neural network combined with causal convolutional attention mechanisms for ECG signal classification from the PhysioNet MIT-BIH Arrhythmia database. Simultaneously, the dataset is balanced by downsampling the majority class and oversampling the minority class using the Synthetic Minority Oversampling Technique (SMOTE), effectively categorizing the five heartbeat types in the test dataset. The experimental results showcase the classifier’s performance, evaluated through accuracy, precision, sensitivity, and F1-score and culminating in an overall accuracy of 99.35%, precision of 96.55%, sensitivity of 96.73%, and an F1-recall of 96.63%, surpassing existing methods. Simultaneously, the application of this innovative data balancing technique significantly addresses the issue of data imbalance. Compared to the data before balancing, there was a significant improvement in accuracy for the S-class and the F-class, with increases of approximately 8% and 13%, respectively.

CTCNet: A CNN Transformer capsule network for sleep stage classification

In this paper, we propose a novel neural network architecture called CTCNet. First, we adopt a multi-scale convolutional neural network (MSCNN) to extract low and high-frequency features, adaptive channel feature recalibration (ACFR) to enhance the model’s sensitivity to important channel features in the feature maps and reduce dependence on irrelevant or redundant features, a multi-scale dilated convolutional block (MSDCB) to capture characteristics of different types among feature channels. Second, we use Transformer to extract global temporal context features. Third, we employ capsule network to capture spatial location relationships among EEG features and refine these features. Besides, the capsule network module is used as our model’s classifier to classify the final results. It is worth noting that our model better solves the problem that previous researches failed to take into account the simultaneous extraction of local features and global temporal context characteristics of EEG signals, and ignored the spatial location relationships between these features. Eventually, we assess our model on three datasets and it achieves better or comparable performance than most state-of-the-art methods.

A Lightweight Network Based on Multi-Scale Asymmetric Convolutional Neural Networks with Attention Mechanism for Ship-Radiated Noise Classification

Ship-radiated noise classification is critical in ocean acoustics. Recently, the feature extraction method combined with time–frequency spectrograms and convolutional neural networks (CNNs) has effectively described the differences between various underwater targets. However, many existing CNNs are challenging to apply to embedded devices because of their high computational costs. This paper introduces a lightweight network based on multi-scale asymmetric CNNs with an attention mechanism (MA-CNN-A) for ship-radiated noise classification. Specifically, according to the multi-resolution analysis relying on the relationship between multi-scale convolution kernels and feature maps, MA-CNN-A can autonomously extract more fine-grained multi-scale features from the time–frequency domain. Meanwhile, the MA-CNN-A maintains its light weight by employing asymmetric convolutions to balance accuracy and efficiency. The number of parameters introduced by the attention mechanism only accounts for 0.02‰ of the model parameters. Experiments on the DeepShip dataset demonstrate that the MA-CNN-A outperforms some state-of-the-art networks with a recognition accuracy of 98.2% and significantly decreases the parameters. Compared with the CNN based on three-scale square convolutions, our method has a 68.1% reduction in parameters with improved recognition accuracy. The results of ablation explorations prove that the improvements benefit from asymmetric convolution, multi-scale block, and attention mechanism. Additionally, MA-CNN-A shows a robust performance against various interferences.

Inter-turn short circuit and demagnetization fault diagnosis of ship PMSM based on multiscale residual dilated CNN and BiLSTM

Inter-turn short circuit (ITSC) and demagnetization of permanent magnet synchronous motors (PMSMs) can lead to serious ship accidents, timely and accurate fault diagnosis of these faults is very important. A multi-signal fusion fault diagnosis method (MD-CNN-BiLSTM) is proposed based on multi-scale residual dilated convolutional neural network (D-CNN) and bidirectional long and short-term memory (BiLSTM) for PMSM fault diagnosis. This method first takes three-phase current and vibration signals as input; uses a three-column parallel CNN structure with different scales to extract both global signal and local feature. A residual connection in the expanded CNN is then used to eliminate the problems of gradient disappearance or explosion; and finally, BiLSTM is used to further extract features and identify the fault. A 2.2 kW permanent magnet synchronous motor was used to build a fault simulation test rig. The motor stator was rewound to simulate the ITSC fault, and different sizes of permanent magnets were replaced to simulate demagnetization fault. ITSC, demagnetization and their coupled faults were simulated under 10 specific motor speeds and loads respectively. The test proved that the diagnostic accuracy of the proposed method was 4.2% higher than that of ordinary CNN and 29.06% higher than that of BiLSTM. It also had the best diagnostic effect under the noise interference of different intensities. It was verified that the proposed method has good noise interference and strong classification ability.

Multiscale 1-DCNN for Damage Localization and Quantification Using Guided Waves With Novel Data Fusion Technique and New Self-Attention Module

This article proposes an innovative damage localization and size quantification method named as MSCNNSAM based on new multiscale convolutional neural network (MSCNN) and novel small-weighted zero-setting self-attentive module (SAM) in carbon fiber reinforced plastic structures. Firstly, an improved piecewise aggregate approximation algorithm (IPAA) is developed to compress the guided wave signal and extract a series of damage indexes (DI). Considering the different effects of the damage location on the different sensing paths, a new method of damage information targeting enhancement and multipath data fusion is proposed. Then, a novel MSCNN architecture is also proposed for the inherent multiscale characteristics of the guided wave signal, which takes the multipath fused data as input and uses regression and classification methods to directly predict the location and size of the damage. Finally, to further improve the performance of the MSCNN, a SAM is proposed to effectively avoid the influence of low-information channel features and improve the damage feature extraction capability of the network. The proposed method is evaluated through experiments on a guided wave testing platform. Experimental results and comprehensive comparison analysis with respect to the state-of-the-art damage localization and quantification methods have demonstrated the superiority of the proposed MSCNNSAM.

Fault Diagnosis Method for Bearing Based on Attention Mechanism and Multi-Scale Convolutional Neural Network

Fault Diagnosis Method for Bearing Based on Attention Mechanism and Multi-Scale Convolutional Neural Network

Research on Multi-scale CNN and Transformer-based Multi-level Multi-classification Method for Images

Research on Multi-scale CNN and Transformer-based Multi-level Multi-classification Method for Images

Method of Underwater Acoustic Signal Denoising Based on Dual-Path Transformer Network

The presence of natural ambient noise interferes with the system for locating and identifying underwater targets. This paper suggests that a Dual-Path Transformation Network (DPTN) reduces ambient noise in underwater acoustic signals. First, the input acoustic signals’ higher-order non-linear features are extracted using a multi-scale convolutional encoder neural network. Second, sub-vectors with the same length are created according to the time dimension from the higher-order non-linear features. The sub-vectors are stitched together to form a three-dimensional tensor. Third, a neural network transformer based on the feed-forward network is constructed. Further, to capture long-term series features and separate the target signal from the noisy signals, the three-dimensional tensor is used as the input of the transformer-based masking network. Finally, overlap-add and transpose are used to obtain discernible target signals. The experimental results verify the effectiveness of the proposed underwater acoustic signal denoising algorithm and demonstrate that the proposed DPRN method can obtain higher output signal-to-noise ratio (SNR) and the scale-invariant signal-to-noise ratio (SI-SNR) compared with the other classical algorithms.

Multiscale 3-D–2-D Mixed CNN and Lightweight Attention-Free Transformer for Hyperspectral and LiDAR Classification

Multiscale 3-D–2-D Mixed CNN and Lightweight Attention-Free Transformer for Hyperspectral and LiDAR Classification

A Residual Multi-Scale Convolutional Neural Network with Transformers for Speech Emotion Recognition

A Residual Multi-Scale Convolutional Neural Network with Transformers for Speech Emotion Recognition

A Hybrid BCI Integrating EEG and Eye-Tracking for Assisting Clinical Communication in Patients With Disorders of Consciousness.

Assessing communication abilities in patients with disorders of consciousness (DOCs) is challenging due to limitations in the behavioral scale. Electroencephalogram-based brain-computer interfaces (BCIs) and eye-tracking for detecting ocular changes can capture mental activities without requiring physical behaviors and thus may be a solution. This study proposes a hybrid BCI that integrates EEG and eye tracking to facilitate communication in patients with DOC. Specifically, the BCI presented a question and two randomly flashing answers (yes/no). The subjects were instructed to focus on an answer. A multimodal target recognition network (MTRN) is proposed to detect P300 potentials and eye-tracking responses (i.e., pupil constriction and gaze) and identify the target in real time. In the MTRN, the dual-stream feature extraction module with two independent multiscale convolutional neural networks extracts multiscale features from multimodal data. Then, the multimodal attention strategy adaptively extracts the most relevant information about the target from multimodal data. Finally, a prototype network is designed as a classifier to facilitate small-sample data classification. Ten healthy individuals, nine DOC patients and one LIS patient were included in this study. All healthy subjects achieved 100% accuracy. Five patients could communicate with our BCI, with 76.1±7.9% accuracy. Among them, two patients who were noncommunicative on the behavioral scale exhibited communication ability via our BCI. Additionally, we assessed the performance of unimodal BCIs and compared MTRNs with other methods. All the results suggested that our BCI can yield more sensitive outcomes than the CRS-R and can serve as a valuable communication tool.

DDoS‐MSCT: A DDoS Attack Detection Method Based on Multiscale Convolution and Transformer

Distributed denial‐of‐service (DDoS) attacks pose a significant threat to network security due to their widespread impact and detrimental consequences. Currently, deep learning methods are widely applied in DDoS anomaly traffic detection. However, they often lack the ability to collectively model both local and global traffic features, which presents challenges in improving performance. In order to provide an effective method for detecting abnormal traffic, this paper proposes a novel network architecture called DDoS‐MSCT, which combines a multiscale convolutional neural network and transformer. The DDoS‐MSCT architecture introduces the DDoS‐MSCT block, which consists of a local feature extraction module (LFEM) and a global feature extraction module (GFEM). The LFEM employs convolutional kernels of different sizes, accompanied by dilated convolutions, with the aim of enhancing the receptive field and capturing multiscale features simultaneously. On the other hand, the GFEM is utilized to capture long‐range dependencies for attending to global features. Furthermore, with the increase in network depth, DDoS‐MSCT facilitates the integration of multiscale local and global contextual information of traffic features, thereby improving detection performance. Our experiments are conducted on the CIC‐DDoS2019 dataset, and also the CIC‐IDS2017 dataset, which is introduced as a supplement to address the issue of sample imbalance. Experimental results on the hybrid dataset show that DDoS‐MSCT achieves accuracy, recall, F1 score, and precision of 99.94%, 99.95%, 99.95%, and 99.97%, respectively. Compared to the state of the art methods, the DDoS‐MSCT model achieves a good performance for detecting the DDoS attack to provide the protecting ability for network security.

Remaining Useful Life Prediction and Uncertainty Quantification for Bearings Based on Cascaded Multiscale Convolutional Neural Network

Remaining Useful Life Prediction and Uncertainty Quantification for Bearings Based on Cascaded Multiscale Convolutional Neural Network

DCMA: faster protein backbone dihedral angle prediction using a dilated convolutional attention-based neural network.

The dihedral angle of the protein backbone can describe the main structure of the protein, which is of great significance for determining the protein structure. Many computational methods have been proposed to predict this critically important protein structure, including deep learning. However, these heavyweight methods require more computational resources, and the training time becomes intolerable. In this article, we introduce a novel lightweight method, named dilated convolution and multi-head attention (DCMA), that predicts protein backbone torsion dihedral angles . DCMA is stacked by five layers of two hybrid inception blocks and one multi-head attention block (I2A1) module. The hybrid inception blocks consisting of multi-scale convolutional neural networks and dilated convolutional neural networks are designed for capturing local and long-range sequence-based features. The multi-head attention block supplementally strengthens this operation. The proposed DCMA is validated on public critical assessment of protein structure prediction (CASP) benchmark datasets. Experimental results show that DCMA obtains better or comparable generalization performance. Compared to best-so-far methods, which are mostly ensemble models and constructed of recurrent neural networks, DCMA is an individual model that is more lightweight and has a shorter training time. The proposed model could be applied as an alternative method for predicting other protein structural features.