Facial expression recognition (FER) continues to be a vibrant research field, driven by the increasing need for its practical applications in areas such as e-learning, healthcare, candidate interview analysis, and mor...
详细信息
Facial expression recognition (FER) continues to be a vibrant research field, driven by the increasing need for its practical applications in areas such as e-learning, healthcare, candidate interview analysis, and more. Most deep learning approaches in supervised FER systems heavily rely on large, labeled datasets. Implementing FER in Convolutional Neural Networks (CNNs) often requires many layers, leading to extended training times and difficulties in finding optimal parameters. This can result in challenges in creating distinct facial expression patterns for classification, leading to poor real-time emotion classification In this paper, we propose a novel approach known as the Deep Semi-supervised Convolutional Sparse autoencoder to address the aforementioned issues and enhance FER performance and prediction accuracy. This approach comprises two parts: (i) Initially, a deep convolutional sparse autoencoder is trained with unlabeled samples of facial expressions. Here, sparsity is introduced in the convolutional block to enforce penalties, focusing on more relevant features for feature representation in the latent space. (ii) A trained encoder with a feature map is connected to a fully connected layer with softmax for final fine-tuning with learned weights and labeled facial expression samples in a semi-supervised approach for emotion classification. This approach was experimented with two benchmark datasets, namely CK + and JAFFE, and achieved significant results of 98.98% and 93.10% accuracy, respectively. The results were analyzed using established state-of-the-art techniques. Additionally, eXplainable AI (XAI) methods like Grad-CAM and image-LIME were employed to interpret the performance and prediction outcomes of the DSCSA model.
Ultrasonic guided wave-based structural health monitoring (UGW-SHM) technology has a highly efficient damage detection capability. However, changes in environmental factors such as temperature can lead to changes in t...
详细信息
Ultrasonic guided wave-based structural health monitoring (UGW-SHM) technology has a highly efficient damage detection capability. However, changes in environmental factors such as temperature can lead to changes in the signal waveform, which greatly affects the damage detection performance. In this paper, a novel unsupervised temperature compensated damage localization method called damage to baseline autoencoder - delay-based probabilistic imaging (DBAE-DPI) method is proposed. Firstly, based on the propagation mechanism of Lamb waves and the unsupervised learning method, the temperature-compensated unsupervised model DBAE is designed. This model does not require any damage samples during the training stage and consists of a path fusion module, an encoder, and a decoder, which fully extract and utilize environmental feature information from all paths to achieve baseline signal reconstruction for the detected environments. Secondly, unlike the existing unsupervised learning models in the field of SHM, the output of DBAE is the reconstructed baseline signal of all paths;based on the output signal and input damage signal, the scattering signal can be computed to extract richer damage information. Thirdly, a temperature-compensated DPI damage imaging method is designed, which is based on the temperature-compensated reconstructed signal of DBAE and combines the time delay principle with the probabilistic imaging algorithm to achieve more accurate damage localization effect under temperature-varying environments. Finally, experiments were conducted on the Open Guided Waves (OGW) temperature compensation benchmark dataset to evaluate the temperature compensation and damage localization performance of the DBAE-DPI method on carbon fiber composite plates. The results show that the method has a better temperature compensation effect and damage localization accuracy than other damage localization methods validated using the OGW dataset (In the temperature range of 20 degrees
The increasing sophistication of network attacks, particularly zero-day threats, underscores the need for robust, unsupervised detection methods. These attacks can flood networks with malicious traffic, overwhelm reso...
详细信息
The increasing sophistication of network attacks, particularly zero-day threats, underscores the need for robust, unsupervised detection methods. These attacks can flood networks with malicious traffic, overwhelm resources, or render services unavailable to legitimate users. Existing machine learning methods for zero-day attack detection typically rely on statistical features of network traffic, such as packet sizes and inter-arrival times. However, traditional approaches that depend on manually labeled data and linear structures often struggle to capture the intricate spatiotemporal correlations crucial for detecting unknown attacks. This paper introduces the Multiscale Temporal Convolutional Recurrent autoencoder (MTCR-AE), an innovative framework designed to detect malicious network traffic by leveraging Multiscale Temporal Convolutional Networks (TCN) and Gated Recurrent Units (GRU). The MTCR-AE model captures both short- and long-range spatiotemporal dependencies while incorporating a temporal attention mechanism to dynamically prioritize critical features. The MTCR-AE operates in an unsupervised manner, eliminating the need for manual data labeling and enhancing its scalability for real-world applications. Experimental evaluations conducted on four benchmark datasets - ISCX-IDS-2012, USTC-TFC-2016, CIRA-CIC-DoHBrw2020, and CICIoT2023 - demonstrate the model's superior performance, achieving an accuracy of 99.69%, precision of 99.63%, recall of 99.69%, and an F1-score of 99.66%. The results highlight the model's capability to deliver state-of-the-art detection performance while maintaining low false positive and false negative rates, offering a scalable and reliable solution for dynamic network environments.
Asphalt concrete (AC) balanced mix design (BMD) is based on the selection of aggregate gradation, component volumetrics, and binder content to control pavement cracking and rutting potential. The Illinois Flexibility ...
详细信息
Asphalt concrete (AC) balanced mix design (BMD) is based on the selection of aggregate gradation, component volumetrics, and binder content to control pavement cracking and rutting potential. The Illinois Flexibility Index Test (I-FIT) and the Hamburg Wheel Tracking Test (HWTT) results, used to predict cracking and rutting potential, respectively, are used in the BMD approach. However, BMD generally relies on a trial-and-error process to identify the aggregate gradation and binder content needed to meet volumetrics and optimize I-FIT and HWTT results. Minimizing or eliminating the trial-and-error process would increase productivity and accuracy. Therefore, this study proposes an autoencoder deep neural network (ADNN) to develop optimized AC mix design alternatives that can meet a prescribed flexibility index (FI) and rut depth (RD). autoencoders are a type of neural network designed for representation learning composed of an encoder and a decoder. The encoder detects a structured pattern in the original input data to create a compressed representation of the AC mix design. The decoder reconstructs the compressed representation. The proposed autoencoder is composed of an encoder of five hidden layers, a latent space of one node, and a five-hidden-layer decoder. Models were created from a database of 5,357 data sets that include mix properties, I-FIT FI, and HWTT RD (after data preprocessing was conducted). An autoencoder was then trained to predict the total binder content, and aggregate gradation based on a target mix type, FI, and RD.
作者:
Yizhe WangDeqing WangLiqun FuSchool of Informatics
and Key Laboratory of Underwater Acoustic Communication and Marine Information Technology of Ministry of EducationXiamen UniversityXiamen 361005China
The long delay spreads and significant Doppler effects of underwater acoustic(UWA)channels make the design of the UWA communication system more *** this paper,we propose a learning-based end-to-end framework for UWA c...
详细信息
The long delay spreads and significant Doppler effects of underwater acoustic(UWA)channels make the design of the UWA communication system more *** this paper,we propose a learning-based end-to-end framework for UWA communications,leveraging a double feature extraction network(DFEN)for data *** DFEN consists of an attentionbased module and a mixer-based module for channel feature extraction and data feature extraction,*** the diverse nature of UWA channels,we propose a stack-network with a two-step training strategy to enhance *** avoiding the use of pilot information,the proposed network can learn data mapping that is robust to UWA *** results show that our proposed algorithm outperforms the baselines by at least 2 dB under bit error rate(BER)10^(−2)on the simulation channel,and surpasses the compared neural network by at least 5 dB under BER 5×10^(−2)on the experiment channels.
The ability to detect anomalies in business processes is crucial for achieving success in business operations. While unsupervised anomaly detection approaches have gained popularity in recent years due to their label-...
详细信息
The ability to detect anomalies in business processes is crucial for achieving success in business operations. While unsupervised anomaly detection approaches have gained popularity in recent years due to their label-free nature, in some cases, a limited number of labelled anomalies can be provided and using them can improve the performance of anomaly detection. To address this issue, we propose a novel framework for anomaly detection that uses a pre-trained autoencoder to extract feature representations of traces. An anomaly score generator based on a multi-layer perceptron is utilized to evaluate the extracted features. The entire framework is trained using a joint loss that ensures the generated anomaly scores satisfy a specific distribution without compromising the autoencoder's ability to reconstruct normal traces. The feature encoder is fine-tuned to provide insights into the cause of anomalies. Additionally, we design a novel technique for calculating anomaly scores to mitigate the effects of varying numbers of potential attribute values. We conduct extensive experiments on both synthetic and real-life logs, and our results demonstrate that our proposed method, WAKE, outperforms state-of-the-art unsupervised deep business process anomaly detection methods by a significant margin. Additionally, it outperforms other weakly supervised anomaly detection methods as well.
Nowadays, studies on indoor localization systems based on wireless systems are increasing widely. Indoor localization is the process of determining the location of objects or people inside a building. Global Navigatio...
详细信息
Nowadays, studies on indoor localization systems based on wireless systems are increasing widely. Indoor localization is the process of determining the location of objects or people inside a building. Global Navigation Satellite System (GPS) signals do not provide sufficient location data indoors because they are interrupted or completely lost in closed areas. For this reason, studies on indoor localization system design with machine learning and deep learning techniques based on Wi-Fi technology are increasing. In this study, we propose a method and training strategy that is entirely based on a Convolutional Neural Network (CNN) and a combined autoencoder that automatically extracts features from Wi-Fi fingerprint samples. In this model, we coupled an autoencoder and a CNN and we trained them simultaneously. Thus, we guarantee that the encoder and the CNN are trained simultaneously. The proposed system was evaluated on the UJIIndoorLoc and Tampere datasets. The experimental results show that the proposed model performs significantly better than the current state-of-the-art methods in terms of location coordinates (x, y) localization. In our study, runtime analysis is also presented to show the real-time performance of the network we proposed.
This paper addresses the challenge of identifying abnormal states in Lithium-ion Battery(LiB)time series *** the energy sector increasingly focuses on integrating distributed energy resources,Virtual Power Plants(VPP)...
详细信息
This paper addresses the challenge of identifying abnormal states in Lithium-ion Battery(LiB)time series *** the energy sector increasingly focuses on integrating distributed energy resources,Virtual Power Plants(VPP)have become a vital new framework for energy *** are key in this context,owing to their high-efficiency energy storage capabilities essential for VPP ***,LiBs are prone to various abnormal states like overcharging,over-discharging,and internal short circuits,which impede power transmission *** methods for detecting such abnormalities in LiB are too broad and lack precision for the dynamic and irregular nature of LiB *** response,we introduce an innovative method:a Long Short-Term Memory(LSTM)autoencoder based on Dynamic Frequency Memory and Correlation Attention(DFMCA-LSTM-AE).This unsupervised,end-to-end approach is specifically designed for dynamically monitoring abnormal states in LiB *** method starts with a Dynamic Frequency Fourier Transform module,which dynamically captures the frequency characteristics of time series data across three scales,incorporating a memory mechanism to reduce overgeneralization of abnormal *** is followed by integrating LSTM into both the encoder and decoder,enabling the model to effectively encode and decode the temporal relationships in the time *** tests on a real-world LiB dataset demonstrate that DFMCA-LSTM-AE outperforms existing models,achieving an average Area Under the Curve(AUC)of 90.73%and an F1 score of 83.83%.These results mark significant improvements over existing models,ranging from 2.4%–45.3%for AUC and 1.6%–28.9%for F1 score,showcasing the model’s enhanced accuracy and reliability in detecting abnormal states in LiB data.
In the distributed sensing method, various sensor arrays encompass both sensor readings and potential spatial information. While increasing the number of sensors can enhance accuracy, an alternative approach to achiev...
详细信息
In the distributed sensing method, various sensor arrays encompass both sensor readings and potential spatial information. While increasing the number of sensors can enhance accuracy, an alternative approach to achieve this lies in extracting spatial knowledge from combinations of horizontal and vertical sensor arrays. By adopting a scaffold-based conceptualization of sensor distribution, this study introduces a novel input preprocessing approach termed the "scaffold-based preprocessing autoencoder" (SPA) to augment distributed sensing methods. This approach leverages autoencoders to extract informative features from diverse combinations of sensor arrays arranged in vertical columns and horizontal planes. The effectiveness of the proposed approach is empirically validated in the posture sensing of a vacuum-powered bellow-shaped fluidic elastomer actuator (FEA), employing nine distributed flexible bending sensors. The results demonstrate that the neural network incorporating features extracted from sensor combinations surpasses the performance of conventional long short-term memory (LSTM) neural networks. Comparative investigations validate the distinct advantages of these arrangements, with horizontal combinations yielding improved estimation accuracy in the X- and Y-axes, and vertical combinations enhancing accuracy in the Z-axis. Collectively, these combinations yield reductions in root mean square error (RMSE) of 9.51-6.93 mm in the X-axis, 6.77-5.43 mm in the Y-axis, and 7.72-4.38 mm in the Z-axis. Besides, the application of SPA to the gated recurrent unit (GRU) and bidirectional LSTM (BiLSTM) models also demonstrated significant improvements. Both models exhibited a substantial reduction in the sum of RMSE in each axis, with decreases of 20% and 27%, respectively.
The separation of P- and S-waves is pivotal in the processing of multicomponent seismic data. The complexity of geological structures often leads to intricate P- and S-wavefields, which poses challenges for identifyin...
详细信息
The separation of P- and S-waves is pivotal in the processing of multicomponent seismic data. The complexity of geological structures often leads to intricate P- and S-wavefields, which poses challenges for identifying and separating waves using conventional signal features, such as the F-K domain or tau-p domain distributions, while data-driven machine learning methods overly rely on the quality of samples. This article proposes a novel approach for P- and S-wave separation in complex geological structures based on a knowledge-guided autoencoder network. First, the existing full waveform inversion (FWI) can obtain relatively accurate knowledge representations of P- and S-waves. A recurrent neural network (RNN) was employed for elastic wave FWI to acquire a knowledge representation of intricate P- and S-waves in complex structures. Subsequently, a dual-branch autoencoder network was constructed based on the obtained knowledge representation of the complex P- and S-waves. One branch was guided by the knowledge representation of P-waves for P-wave separation, whereas the other branch was guided by the knowledge representation of S-waves for S-wave separation. Finally, a comprehensive autoencoder network architecture was devised that incorporates waveform reconstruction loss, P-wave knowledge guidance loss, and S-wave knowledge guidance loss for effective P- and S-wave separation. Theoretical analyses and numerical simulations were performed, and they demonstrated the effectiveness of the proposed method for achieving P- and S-wave separation in complex geological structures.
暂无评论