Multi-feature Extraction Research Articles

Pulse-diagnosis-inspired multi-feature extraction deep network for short-term electricity load forecasting

Forecasting short-term electricity load (STEL) is a very important but challenging task by the fact that the series dynamic change involves in multiple patterns, such as long short-term cyclical, nonlinear abrupt, nonlinear gradual, and linear monotonic patterns, while interactions of the above patterns are unknown and variable. In addition, most existing deep forecasting models only focus on the data-layer fitting, and inherit unexplained flowchart and domain knowledge lacking. Motivated by that traditional Chinese medicine (TCM) practitioner analyzes pulse signal coupled with multiple patterns, i.e., pulse strength, thickness, length, rate, etc. and further diagnoses diseases at the extraction-feature level, this paper develops a pulse-diagnosis-inspired multi-feature extraction deep network with three following components. The sense block mimics that the TCM practitioner applies fingers to feel and extract multiple pulse features, the comprehension block mimics that the TCM practitioner integrates the above pulse features and hence grasps core abnormal features, and the judgement block mimics that the TCM practitioner deduces patient’s diseases based on core abnormal features. Specifically, the first block is applied to intelligently extract multiple pattern features based on the domain knowledge, the second block is used to fuse the above features and obtain core features via the gate mechanism, and the third block is adopted to make final forecasts for STEL. The proposed network learns the pulse diagnosis in both flowchart (layer by layer) and function (three block) aspects, producing explained flowchart and clear block function. Experimental results under two real-world electricity load series with different sampling intervals (30 and 60 min) show that the proposed deep network has higher accuracy and stability than 12 baselines, and improves the root mean absolute percent error by averages of 42.9% and 55.8%, respectively. Thus, the proposed deep network provides reliable electricity evidence for power sector scheduling.

Leakage Detection in Water Distribution Networks Based on Multi-Feature Extraction from High-Frequency Pressure Data

Leakage detection is an important task to ensure the operational safety of water distribution networks. Leakage characteristic extraction based on high-frequency data has been widely used for leakage detection in experimental networks. However, the accuracy of single-feature-based methods is limited by the interference of background pressure fluctuations in networks. In addition, the setting of leakage diagnosis thresholds has been insufficiently studied, but influences leakage detection performance greatly. Hence, a new method of leakage detection is proposed based on multi-feature extraction. The multi-features of leakage are composed of instantaneous characteristics (ICs) and trend characteristics (TCs), which constitute comprehensive leakage information. The levels of the instantaneous and trend pressure drops in background pressure fluctuations in network environments are quantified for the setting of leakage diagnosis thresholds. In addition, ICs and TCs are used for leakage degree prediction. The proposed method was applied to an experimental network. Compared with the single-feature-based method and the cumulative sum (CUSUM) method, the proposed method achieved increases of 6.01% and 13.66% in F-Scores, respectively, and showed better adaptability to background pressure fluctuations in complex network environments.

일변량 시계열의 다중 특징 추출을 통한 시계열 예측 모델 설명

최근 다양한 산업 분야에 기계학습 기술들이 적용됨에 따라, 개발된 AI 모델들의 행동을 이해하고 효과적으로 관리하기 위한 XAI(eXplainable Artificial Intelligence)의 중요성이 강조되고 있다. 시계열 모델의 경우에도 XAI 이슈가 중요하지만, 만약 특징의 수가 충분하지 않은 경우 XAI 기법들의 적용이 용이하지 않다. 이를 위해, 본 논문에서는 일변량 시계열로부터 설명 가능한 모델을 확보하기 위한 다중 특징 추출 방법을 제안한다. 먼저, 시계열 분해 및 특징 공학 라이브러리를 이용하여 일변량 시계열로부터 다변량 시계열을 추출한다. 다음으로 기계학습 알고리즘을 이용하여 다변량 시계열로부터 예측 모델을 생성한다. 마지막으로 XAI 기법을 이용하여 생성된 모델에 대한 설명결과를 출력한다. 이와 관련하여 한국전력공사의 배전선로 부하데이터를 이용한 실험을 통해 제안 방법을 검증한다.

A Hybrid Attention Network for Malware Detection Based on Multi-Feature Aligned and Fusion

With the widespread use of computers, the amount of malware has increased exponentially. Since dynamic detection is costly in both time and resources, most existing malware detection methods are based on static features. However, existing static methods mainly rely on single feature types of malware, while few pay attention to multi-feature fusion. This paper presents a novel multi-feature extraction and fusion method to effectively detect malware variants by combining binary and opcode features. We propose a stacked convolutional network to capture the temporal and discontinuity information in the function call of the binary file from malware. Additionally, we adopt the triangular attention algorithm to extract code-level features from assembly code. Additionally, these two extracted features are aligned and fused by the cross-attention, which could provide a stable feature representation. We evaluate our method on two different datasets. It achieves an accuracy of 0.9954 on the Kaggle Malware Classification dataset and an accuracy of 0.9544 on a large real-world dataset. To optimize our detection model, we conduct in-depth discussions on different feature extractors and multi-feature fusion strategies. Moreover, a visualized attention module in our model is provided to explain its superiority in the opcode feature extraction. An experimental analysis is performed against five baseline deep learning models and five state-of-the-art malware detection models, which reveals that our strategy outperforms competing approaches in all evaluation circumstances.

Micro-Expression Spotting Based on a Short-Duration Prior and Multi-Stage Feature Extraction

When micro-expressions are mixed with normal or macro-expressions, it becomes increasingly challenging to spot them in long videos. Aiming at the specific time prior of micro-expressions (MEs), an ME spotting network called AEM-Net (adaptive enhanced ME detection network) is proposed. This paper is an extension of the conference paper presented at the Chinese Conference on Biometric Recognition (CCBR). The network improves spotting performance in the following five aspects. Firstly, a multi-stage channel feature extraction module is constructed to extract the features at different depths. Then, an attention spatial-temporal module is leveraged to obtain salient and discriminative micro-expression segments while suppressing the generation of excessively long or short suggestions. Thirdly, a ME-NMS (non-maximum suppression) network is developed to reduce redundancy and decision errors. Fourthly, a multi-scale feature fusion module is introduced to fuse up-sampling features of high-level maps and fine-grained information, which obtains meaningful information on feature distribution and contributes to a good representation of MEs. Finally, two spotting mechanisms named anchor-based and anchor free were integrated to get final spotting. Extensive experiments were conducted on prevalent CAS(ME)2 and the SAMM-Long ME databases to evaluate the spotting performance. The results show that the AEM-Net achieves competitive performance, outperforming other state-of-the-art methods.

Quantitative identification of damage in composite structures using sparse sensor arrays and multi-domain-feature fusion of guided waves

Damage detection techniques using Lamb waves have shown excellent capabilities in the diagnosis of composite structures. However, structural health monitoring of composite structures is challenging, especially for damage classification. This study proposes a machine learning-based method with a sparse sensor array to achieve quantitative classification of the damage location and severity on a composite plate. First, multi features extraction is used to construct a support vector machine (SVM) damage localization model. Second, optimal path extraction combined with principal component analysis (PCA) is used to construct an SVM model for classification. To reduce the operational burden of structures, the sparse array is employed. To improve the damage classification accuracy, Fisher clustering is proposed to extract the optimal detection path. Then, PCA is used to achieve data fusion. Experimental results on a glass fiber-reinforced epoxy composite laminate plate demonstrate that the proposed technique can accurately locate and classify the quantitative artificial damage.

IAVPs-ResBi: Identifying antiviral peptides by using deep residual network and bidirectional gated recurrent unit.

Human history is also the history of the fight against viral diseases. From the eradication of viruses to coexistence, advances in biomedicine have led to a more objective understanding of viruses and a corresponding increase in the tools and methods to combat them. More recently, antiviral peptides (AVPs) have been discovered, which due to their superior advantages, have achieved great impact as antiviral drugs. Therefore, it is very necessary to develop a prediction model to accurately identify AVPs. In this paper, we develop the iAVPs-ResBi model using k-spaced amino acid pairs (KSAAP), encoding based on grouped weight (EBGW), enhanced grouped amino acid composition (EGAAC) based on the N5C5 sequence, composition, transition and distribution (CTD) based on physicochemical properties for multi-feature extraction. Then we adopt bidirectional long short-term memory (BiLSTM) to fuse features for obtaining the most differentiated information from multiple original feature sets. Finally, the deep model is built by combining improved residual network and bidirectional gated recurrent unit (BiGRU) to perform classification. The results obtained are better than those of the existing methods, and the accuracies are 95.07, 98.07, 94.29 and 97.50% on the four datasets, which show that iAVPs-ResBi can be used as an effective tool for the identification of antiviral peptides. The datasets and codes are freely available at https://github.com/yunyunliang88/iAVPs-ResBi.

Diagnosis of Parkinson’s disease genes using LSTM and MLP based multi-feature extraction methods

Diagnosis of Parkinson’s disease genes using LSTM and MLP based multi-feature extraction methods

Diagnosis of Parkinson's disease genes using LSTM and MLP-based multi-feature extraction methods

Diagnosis of Parkinson's disease genes using LSTM and MLP-based multi-feature extraction methods

An Early Warning Model for Intelligent Operation of Power Engineering based on Kalman Filter Algorithm

The accurate early warning of intelligent operation of power engineering can find the abnormal operation of substation equipment in time and ensure the safe operation of substation equipment. Thus, an early warning model for intelligent operation of power engineering based on Kalman filter algorithm is constructed. In this model, the noise elimination method of substation equipment inspection image based on particle resampling filter algorithm is introduced. After removing the noise information of operation situation inspection image of substation equipment, the gradient direction histogram feature, lab color space feature and edge contour feature in the image are extracted by the multi-feature extraction method for intelligent operation of power engineering based on multi-feature fusion. These features are combined to form the feature description set of equipment operation situation. The feature description set is used as the identification attribute set of the anomaly identification and early warning model for intelligent operation of electric power engineering based on Kalman filter algorithm to complete the anomaly identification and early warning of equipment operation situation. The test shows that when the model is used to observe the temperature change trend of the top layer of the transformer, the temperature error is very small, and the early warning accuracy for the abnormal temperature of the top layer of the transformer is very high, so the abnormal operation of the substation equipment can be found in time.

State of Health Estimation of Lithium-Ion Batteries Using a Multi-Feature-Extraction Strategy and PSO-NARXNN

The lithium-ion battery state of health (SOH) estimation is critical for maintaining reliable and safe working conditions for electric vehicles (EVs). However, accurate and robust SOH estimation remains a significant challenge. This paper proposes a multi-feature extraction strategy and particle swarm optimization-nonlinear autoregressive with exogenous input neural network (PSO-NARXNN) for accurate and robust SOH estimation. First, eight health features (HFs) are extracted from partial voltage, capacity, differential temperature (DT), and incremental capacity (IC) curves. Then, qualitative and quantitative analyses are used to evaluate the selected HFs. Second, the PSO algorithm is adopted to optimize the hyperparameters of NARXNN, including input delays, feedback delays, and the number of hidden neurons. Third, to verify the effectiveness of the multi-feature extraction strategy, the SOH estimators based on a single feature and fusion feature are comprehensively compared. To verify the effectiveness of the proposed PSO-NARXNN, a simple three-layer backpropagation neural network (BPNN) and a conventional NARXNN are built for comparison based on the Oxford aging dataset. The experimental results demonstrate that the proposed method has higher accuracy and stronger robustness for SOH estimation, where the average mean absolute error (MAE) and root mean square error (RMSE) are 0.47% and 0.56%, respectively.

M³LVI: a multi-feature, multi-metric, multi-loop, LiDAR-visual-inertial odometry via smoothing and mapping

PurposeThe purpose of this paper is to propose a multi-feature, multi-metric and multi-loop tightly coupled LiDAR-visual-inertial odometry, M3LVI, for high-accuracy and robust state estimation and mapping.Design/methodology/approachM3LVI is built atop a factor graph and composed of two subsystems, a LiDAR-inertial system (LIS) and a visual-inertial system (VIS). LIS implements multi-feature extraction on point cloud, and then multi-metric transformation estimation is implemented to realize LiDAR odometry. LiDAR-enhanced images and IMU pre-integration have been used in VIS to realize visual odometry, providing a reliable initial guess for LIS matching module. Location recognition is performed by a dual loop module combined with Bag of Words and LiDAR-Iris to correct accumulated drift. M³LVI also functions properly when one of the subsystems failed, which greatly increases the robustness in degraded environments.FindingsQuantitative experiments were conducted on the KITTI data set and the campus data set to evaluate the M3LVI. The experimental results show the algorithm has higher pose estimation accuracy than existing methods.Practical implicationsThe proposed method can greatly improve the positioning and mapping accuracy of AGV, and has an important impact on AGV material distribution, which is one of the most important applications of industrial robots.Originality/valueM3LVI divides the original point cloud into six types, and uses multi-metric transformation estimation to estimate the state of robot and adopts factor graph optimization model to optimize the state estimation, which improves the accuracy of pose estimation. When one subsystem fails, the other system can complete the positioning work independently, which greatly increases the robustness in degraded environments.

CapsGaNet: Deep Neural Network Based on Capsule and GRU for Human Activity Recognition

The advances in deep learning with the ability to automatically extract advanced features have achieved a bright prospect for human activity recognition (HAR). However, the traditional HAR methods still have the deficiencies of incomplete feature extraction, which may lead to incorrect recognition results. To resolve the above problem, a novel framework for spatiotemporal multi-feature extraction on HAR called CapsGaNet is propounded, which is based on capsule and gated recurrent units (GRU) with attention mechanisms. The proposed framework involves a spatial feature extraction layer consisting of capsule blocks, a temporal feature extraction layer consisting of GRU with attention mechanisms, and an output layer. At the same time, considering the actual demands for recognizing aggressive activities in some specific scenarios like smart prison, we constructed a daily and aggressive activity dataset (DAAD). Moreover, based on the acceleration characteristics of aggressive activity, a threshold-based approach for aggressive activity detection is propounded to meet the needs of high real-time and low computational complexity in prison scenarios. The experiments are performed on the wireless sensor data mining (WISDM) dataset and the DAAD dataset, and the results verify that the propounded CapsGaNet could effectually improve the recognition accuracy. The proposed threshold-based approach for aggressive activity detection provides a more effective HAR way by using smart sensor devices in smart prison scenarios.

Multi-scale permutation Lempel-Ziv complexity and its application in feature extraction for Ship-radiated noise

Permutation Lempel-Ziv complexity (PLZC) is a recently proposed method for analyzing signal complexity. However, PLZC only characterizes the signal complexity from single scale and has certain limitations. In order to overcome these shortcomings and improve the performance of feature extraction for underwater acoustic signal, this paper introduced coarse graining operation, proposed the multi-scale permutation Lempel-Ziv complexity (MPLZC), and proposed an automatic hybrid multi-feature extraction method for ship-radiated noise signal (S-S) based on multi-scale Lempel-Ziv complexity (MLZC), multi-scale permutation entropy (MPE) and MPLZC. The results of simulation and realistic experiments show that MPLZC can better reflect the change of signal complexity in detecting the dynamic change of signals, and more effectively distinguish white noise, pink noise and blue noise than MPE and MLZC; compared with the three feature extraction methods based on MLZC, MPE and MPLZC respectively, the proposed method has the highest recognition rates of six S-Ss under the same number of features, and the recognition rate reaches 100% when the number of features is 5; the introduction of MPLZC significantly improves the performance for ship-radiated noise signal of the automatic hybrid multi-feature extraction method. It is indicated that the proposed method, as a new underwater acoustic technology, is valid in other underwater acoustic signals.

Collaborative multi-feature extraction and scale-aware semantic information mining for medical image segmentation

Objective. In recent years, methods based on U-shaped structure and skip connection have achieved remarkable results in many medical semantic segmentation tasks. However, the information integration capability of this structure is still limited due to the incompatibility of feature maps of encoding and decoding stages at corresponding levels and lack of extraction of valid information in the final stage of encoding. This structural defect is particularly obvious in segmentation tasks with non-obvious, small and blurred-edge targets. Our objective is to design a novel segmentation network to solve the above problems. Approach. The segmentation network named Global Context-Aware Network is mainly designed by inserting a Multi-feature Collaboration Adaptation (MCA) module, a Scale-Aware Mining (SAM) module and an Edge-enhanced Pixel Intensity Mapping (Edge-PIM) into the U-shaped structure. Firstly, the MCA module can integrate information from all encoding stages and then effectively acts on the decoding stages, solving the problem of information loss during downsampling and pooling. Secondly, the SAM module can further mine information from the encoded high-level features to enrich the information passed to the decoding stage. Thirdly, Edge-PIM can further refine the segmentation results by edge enhancement. Main results. We newly collect Magnetic Resonance Imaging of Colorectal Cancer Liver Metastases (MRI-CRLM) dataset in different imaging sequences with non-obvious, small and blurred-edge liver metastases. Our method performs well on the MRI-CRLM dataset and the publicly available ISIC-2018 dataset, outperforming state-of-the-art methods such as CPFNet on multiple metrics after boxplot analysis, indicating that it can perform well on a wide range of medical image segmentation tasks. Significance. The proposed method solves the problem mentioned above and improved segmentation accuracy for non-obvious, small and blurred-edge targets. Meanwhile, the proposed visualization method Edge-PIM can make the edge more prominent, which can assist medical radiologists in their research work well.

Internet of things intrusion detection model and algorithm based on cloud computing and multi-feature extraction extreme learning machine

With the rapid development of the Internet of Things (IoT), there are several challenges pertaining to security in IoT applications. Compared with the characteristics of the traditional Internet, the IoT has many problems, such as large assets, complex and diverse structures, and lack of computing resources. Traditional network intrusion detection systems cannot meet the security needs of IoT applications. In view of this situation, this study applies cloud computing and machine learning to the intrusion detection system of IoT to improve detection performance. Usually, traditional intrusion detection algorithms require considerable time for training, and these intrusion detection algorithms are not suitable for cloud computing due to the limited computing power and storage capacity of cloud nodes; therefore, it is necessary to study intrusion detection algorithms with low weights, short training time, and high detection accuracy for deployment and application on cloud nodes. An appropriate classification algorithm is a primary factor for deploying cloud computing intrusion prevention systems and a prerequisite for the system to respond to intrusion and reduce intrusion threats. This paper discusses the problems related to IoT intrusion prevention in cloud computing environments. Based on the analysis of cloud computing security threats, this study extensively explores IoT intrusion detection, cloud node monitoring, and intrusion response in cloud computing environments by using cloud computing, an improved extreme learning machine, and other methods. We use the Multi-Feature Extraction Extreme Learning Machine (MFE-ELM) algorithm for cloud computing, which adds a multi-feature extraction process to cloud servers, and use the deployed MFE-ELM algorithm on cloud nodes to detect and discover network intrusions to cloud nodes. In our simulation experiments, a classical dataset for intrusion detection is selected as a test, and test steps such as data preprocessing, feature engineering, model training, and result analysis are performed. The experimental results show that the proposed algorithm can effectively detect and identify most network data packets with good model performance and achieve efficient intrusion detection for heterogeneous data of the IoT from cloud nodes. Furthermore, it can enable the cloud server to discover nodes with serious security threats in the cloud cluster in real time, so that further security protection measures can be taken to obtain the optimal intrusion response strategy for the cloud cluster.

Detection and Classification of MRI Brain Tumors using S3-DRLSTM Based Deep Learning Model

Developing an automated brain tumor diagnosis system is a highly challenging task in current days, due to the complex structure of nervous system. The Magnetic Resonance Imaging (MRIs) are extensively used by the medical experts for earlier disease identification and diagnosis. In the conventional works, the different types of medical image processing techniques are developed for designing an automated tumor detection system. Still, it remains with the problems of reduced learning rate, complexity in mathematical operations, and high time consumption for training. Therefore, the proposed work intends to implement a novel segmentation-based classification system for developing an automated brain tumor detection system. In this framework, a Convoluted Gaussian Filtering (CGF) technique is used for normalizing the medical images by eliminating the noise artifacts. Then, the Sparse Space Segmentation (S3) algorithm is implemented for segmenting the pre-processed image into the non-overlapping regions. Moreover, the multi-feature extraction model is used for extracting the contrast, correlation, mean, and entropy features from the segmented portions. The Deep Recurrent Long-Short Term Memory (DRLSTM) technique is utilized for predicting the classified label as normal of disease affected. During results analysis, the performance of the proposed system is tested and compared by using various evaluation measures.

Multi-Feature Extraction-Based Defect Recognition of Foundation Pile under Layered Soil Condition Using Convolutional Neural Network

If the layer of soil surrounding a pile is not taken into account during the engineering detection process, the velocity-time curve might show asymptotic diameter shrinkage or diameter expanding features, which would alter the interpretation of the test findings. In this study, we suggest combining multi-feature extraction and a convolutional neural network (CNN) to increase accuracy in pile defect recognition for layered soil conditions and traditional deep learning flaws. First, numerical simulations are run to create velocity–time curves for foundation piles under layered soil conditions. Then, the data are extracted from three dimensions: time domain, frequency domain, and time-frequency domain, respectively, and fused into a set of feature vectors. Finally, a foundation pile defect identification model combining multi-scale features and CNN is established. The findings demonstrate that the CNN model has 97.8% accuracy while the PNN has 28.6% accuracy, demonstrating that the approach is very reliable.

MF-GCN-LSTM: a cloud-edge distributed framework for key positions prediction in grid projects

In this article, we solve the key positions prediction problem of engineering projects in smart grid, which pays more attention to the spatial-temporal distribution of projects. Many studies show that the projects are affected by multi-dimensional features such as time, space, correlation etc. However, few work can accurately predict the key positions of projects based on multi-dimensional features. In order to solve this problem, we propose the idea of multi-feature extraction, and make use of the real-world records trace to conduct multi-dimensional modeling. Then we introduce a multi-dimensional features extraction model: Multi-Feature-based GCN-LSTM (MF-GCN-LSTM) to take the effect of time, space and correlation for predicting the key positions of projects. Experiments on different datasets with various project types have proved that our model can complete the key positions prediction task efficiently. Compared with the other traditional method and non-linear models, our model shows higher prediction accuracy and robustness. Moreover, we show that the whole prediction framework MF-GCN-LSTM can be split and deployed in a distributed manner to accelerate the inference of the model under the cloud edge system.

An Unsupervised Multi-Dimensional Representation Learning Model for Short-Term Electrical Load Forecasting

The intelligent electrical power system is a comprehensive symmetrical system that controls the power supply and power consumption. As a basis for intelligent power supply control, load demand forecasting in power system operation management has attracted considerable research attention in energy management. In this study, we proposed a novel unsupervised multi-dimensional feature learning forecasting model, named MultiDBN-T, based on a deep belief network and transformer encoder to accurately forecast short-term power load demand and implement power generation planning and scheduling. In the model, the first layer (pre-DBN), based on a deep belief network, was designed to perform unsupervised multi-feature extraction feature learning on the data, and strongly coupled features between multiple independent observable variables were obtained. Next, the encoder layer (D-TEncoder), based on multi-head self-attention, was used to learn the coupled features between various locations, times, or time periods in historical data. The feature embedding of the original multivariate data was performed after the hidden variable relationship was determined. Finally, short-term power load forecasting was conducted. Experimental comparison and analysis of various sequence learning algorithms revealed that the forecasting results of MultiDBN-T were the best, and its mean absolute percentage error and root mean square error were improved by more than 40% on average compared with other algorithms. The effectiveness and accuracy of the model were experimentally verified.