Effective Feature Extraction Research Articles

DANTD: A Deep Abnormal Network Traffic Detection Model for Security of Industrial Internet of Things Using High-Order Features

With the development of blockchain, artificial intelligence and data mining technology, abnormal network traffic data has become easy to obtain. The traffic detection model detects the traffic patterns in the network to find abnormal traffic that does not conform to the normal traffic law, which has great security significance for Industrial Internet of Things (IIoT) networks and devices in real scenarios. However, previous abnormal detection models rely on expert experience and cannot cope with real-time changes in IIoT scenarios. The manual features cannot be sufficiently representative and adaptive. Moreover, there are few abnormal traffic data in real scenarios, which makes the model unable to fully learn the potential distribution in abnormal data. Therefore, in this work, we propose a deep abnormal network traffic detection model (DANTD) for security of IIoT using high-order features and novel data augmentation strategies. The DANTD model first adopts a deep convolutional autoencoder to extract effective high-order features to make it more representative. Then the DANTD model uses generative adversarial networks as data augmentation strategies to enrich the abnormal data, so that the model can fully consider the information of the data distribution. Comprehensive experiments on real IIoT datasets validate the effectiveness of the DANTD model.

Multi-scale cross-layer fusion and center position network for pedestrian detection

Pedestrian detection has made breakthroughs after the rise of convolutional neural networks. However, it faces some challenging problems, including dataset difference, small pedestrian targets and occlusions between pedestrians. To deal with these problems, we propose a novel convolutional network architecture, named multi-scale cross-layer fusion and center position network (MCF-CP-NET). A new backbone unit is designed to introduce channel-wise attention into the improved aggregated residual transformations for effective feature extraction. We select suitable anchor setting for pedestrian detection datasets to tackle the problem of dataset difference. A feature pyramid sub-network with cross-layer fusion is developed for better detection of small pedestrian targets, where cross-layer connections are used to reduce the information loss and low-level marginal feature dissipation and better fuse low- and high-level features. We add a center position branch into the localization regression sub-network in MCF-CP-NET to better detect occluded pedestrians, which predicts the centrality index of the localization box to obtain the center score, and further optimizes the score of non-maximum suppression. Experiments show that the average precision and recall of MCF-CP-NET are separately improved by 1.2% and 0.7% on the person class of Pascal VOC2007 dataset and 1.6% and 0.1% on WiderPerson dataset, in comparison with the state-of-the-art.

Fuzzy Genetic Particle Swarm Optimization Convolution Neural Network Based On Oral Cancer Identification System

Oral cancer is the eighth most common type of cancer in the world. Every year, 130,000 people in India die from mouth cancer. Getting a diagnosis from a clinical exam by skilled doctors and a biopsy takes time. When a problem is found early, it is always easier to treat. The primary goal of this work is to recognise disease-affected oral regions in a given oral image and classify the oral cancer disorder. This study employs unique Deep Learning algorithms to detect the location of disease-affected oral areas. This work employs the most effective feature extraction techniques, including appearance and patter-based features. Following feature extraction, the Bee Pulse Couple Neural Network (BeePCNN) algorithm is used to choose the best feature. Finally, Deep Learning is used to classify these attributes. An innovative FGPSOCNN reduces the computational complexity of CNN. On an additional real-time data set from Arthi Scan Hospital, a secondary evaluation is conducted. The experimental results indicate that the innovative FGPSOCNN performs better than existing methods.

AlexNet-Based Feature Extraction for Cassava Classification: A Machine Learning Approach

Cassava, a significant crop in Africa, Asia, and South America, is a staple food for millions. However, classifying cassava species using conventional color, texture, and shape features is inefficient, as cassava leaves exhibit similarities across different types, including toxic and non-toxic varieties. This research aims to overcome the limitations of traditional classification methods by employing deep learning techniques with pre-trained AlexNet as the feature extractor to accurately classify four types of cassava: Gajah, Manggu, Kapok, and Beracun. The dataset was collected from local farms in Lamongan Indonesia. To collect images with agricultural research experts, the dataset consists of 1,400 images, and each type of cassava has 350 images. Three fully connected (FC) layers were utilized for feature extraction, namely fc6, fc7, and fc8. The classifiers employed were support vector machine (SVM), k-nearest neighbors (KNN), and Naive Bayes. The study demonstrated that the most effective feature extraction layer was fc6, achieving an accuracy of 90.7% with SVM. SVM outperformed KNN and Naive Bayes, exhibiting an accuracy of 90.7%, sensitivity of 83.5%, specificity of 93.7%, and F1-score of 83.5%. This research successfully addressed the challenges in classifying cassava species by leveraging deep learning and machine learning methods, specifically with SVM and the fc6 layer of AlexNet. The proposed approach holds promise for enhancing plant classification techniques, benefiting researchers, farmers, and environmentalists in plant species identification, ecosystem monitoring, and agricultural management.

Object Detection Based on an Improved YOLOv7 Model for Unmanned Aerial-Vehicle Patrol Tasks in Controlled Areas

When working with objects on a smaller scale, higher detection accuracy and faster detection speed are desirable features. Researchers aim to endow drones with these attributes in order to improve performance when patrolling in controlled areas for object detection. In this paper, we propose an improved YOLOv7 model. By incorporating the variability attention module into the backbone network of the original model, the association between distant pixels is increased, resulting in more effective feature extraction and, thus, improved model detection accuracy. By improving the original network model with deformable convolution modules and depthwise separable convolution modules, the model enhances the semantic information extraction of small objects and reduces the number of model parameters to a certain extent. Pretraining and fine-tuning techniques are used for training, and the model is retrained on the VisDrone2019 dataset. Using the VisDrone2019 dataset, the improved model achieves an mAP50 of 52.3% on the validation set. Through the visual comparative analysis of the detection results in our validation set, we find that the model shows a significant improvement in detecting small objects compared with previous iterations.

Fault detection in complex mechatronic systems by a hierarchical graph convolution attention network based on causal paths

Fault detection in mechatronic systems is crucial for their maintainability and safety. However, the systems monitoring variables are often abundant, with intricate connections. It is difficult to characterize their relationships and to extract effective features. In this paper, a hierarchical graph convolution attention network based on causal paths (HGCAN) is developed to improve the performance and interpretability of data-driven fault detection models in complex mechatronic systems. A hybrid causal discovery algorithm is introduced to discover the inherit causality among monitoring variables. The causal paths that sequentially connect the cause-effect variables serve as the reception fields to extract features using multiscale convolution. Different levels of the features are aggregated based on a hierarchical attention mechanism, which assigns adaptive weights considering the varied feature importance. To verify the effectiveness of the proposed method, a dataset of real high-speed train braking systems is considered. Experimental results demonstrate promising performance improvement of the proposed method, and analysis on interpretability indicates its potential to facilitate practical decision-making.

DSAA-YOLO: UAV remote sensing small target recognition algorithm for YOLOV7 based on dense residual super-resolution and anchor frame adaptive regression strategy

The challenges posed by high pixel resolution and complex backgrounds in UAV remote sensing images have hindered the effective feature extraction and precise bounding box regression for small targets. In response to these challenges, numerous detection methods based on deep learning have emerged in recent years. Despite their advancements, these methods have not fully addressed the demand for accurate identification of small targets in remote sensing images. This paper introduces DSAA-YOLO, a novel algorithm designed for small target detection in UAV remote sensing images. Firstly, a new data augmentation strategy, termed Super Resolution Data Augment (SRDA), is proposed, which integrates the concept of image super-resolution to enrich the dataset while preserving data quality. Furthermore, a Dense Residual-based Super-Resolution module (DRSR) is introduced to enhance the resolution of small targets that have undergone quality degradation due to transformations. Subsequently, an Information Alignment Feature Enhancement Module (IAFE) is proposed to maximize the extraction of original features from the image. Finally, based on the improved Multi-Objective Grey Wolf Optimization (MOGWO), a novel dynamic anchor regression strategy termed Multi-Object Golf Dynamic Anchor (MGDA) is devised to generate more precise bounding boxes. The proposed DSAA-YOLO algorithm demonstrates significant improvements over current state-of-the-art methods in terms of widely recognized metrics including mAP, and AP50 on the VisDrone dataset.

Group-Fusion One-Dimensional Convolutional Neural Network for Ballistic Target High-Resolution Range Profile Recognition with Layer-Wise Auxiliary Classifiers

Ballistic missile defense systems require accurate target recognition technology. Effective feature extraction is crucial for this purpose. The deep convolutional neural network (CNN) has proven to be an effective method for recognizing high-resolution range profiles (HRRPs) of ballistic targets. It excels in perceiving local features and extracting robust features. However, the standard CNN's fully connected manner results in high computational complexity, which is unsuitable for deployment in real-time missile defense systems with stringent performance requirements. To address the issue of computational complexity in HRRP recognition based on the standard one-dimensional CNN (1DCNN), we propose a lightweight network called group-fusion 1DCNN with layer-wise auxiliary classifiers (GFAC-1DCNN). GFAC-1DCNN employs group convolution (G-Conv) instead of standard convolution to effectively reduce model complexity. Simply using G-Conv, however, may decrease model recognition accuracy due to the lack of information flow between feature maps generated by each G-Conv. To overcome this limitation, we introduce a linear fusion layer to combine the output features of G-Convs, thereby improving recognition accuracy. Additionally, besides the main classifier at the deepest layer, we construct layer-wise auxiliary classifiers for different hierarchical features. The results from all classifiers are then fused for comprehensive target recognition. Extensive experiments demonstrate that GFAC-1DCNN with such simple and effective techniques achieves higher overall testing accuracy than state-of-the-art ballistic target HRRP recognition models, while significantly reducing model complexity. It also exhibits a higher recall rate for warhead recognition compared to other methods. Based on these compelling results, we believe this work is valuable in reducing workload and enhancing missile interception rates in missile defense systems.

SERS and machine learning based effective feature extraction for detection and identification of amphetamine analogs

Surface-enhanced Raman spectroscopy (SERS) is extensively researched in diverse disciplines due to its sensitivity and non-destructive nature. It is particularly considered a potential and promising technology for rapid on-site screening in drug detection. In this investigation, a technique was developed for fabricating nanocrystals of Ag@Au SNCs. Ag@Au SNCs, as the basic material of SERS, can detect amphetamine at concentrations as low as 1 μg/mL. The Ag@Au SNCs exhibits a strong surface plasmon resonance effect, which amplifies molecular signals. The SERS spectra of ten substances, including amphetamine and its analogs, showed a strong peak signal. To establish a qualitative distinction, we examined the Raman spectra and conducted density functional theory (DFT) calculations on the ten aforementioned species. The DFT calculation enabled us to determine the vibrational frequency and assign normal modes, thereby facilitating the qualitative differentiation of amphetamines and its analogs. Furthermore, the SERS spectrum of the ten mentioned substances was analysed using the support vector machine learning algorithm, which yielded a discrimination accuracy of 98.0 %.

Sign Language Recognition with Multimodal Sensors and Deep Learning Methods

Sign language recognition is essential in hearing-impaired people’s communication. Wearable data gloves and computer vision are partially complementary solutions. However, sign language recognition using a general monocular camera suffers from occlusion and recognition accuracy issues. In this research, we aim to improve accuracy through data fusion of 2-axis bending sensors and computer vision. We obtain the hand key point information of sign language movements captured by a monocular RGB camera and use key points to calculate hand joint angles. The system achieves higher recognition accuracy by fusing multimodal data of the skeleton, joint angles, and finger curvature. In order to effectively fuse data, we spliced multimodal data and used CNN-BiLSTM to extract effective features for sign language recognition. CNN is a method that can learn spatial information, and BiLSTM can learn time series data. We built a data collection system with bending sensor data gloves and cameras. A dataset was collected that contains 32 Japanese sign language movements of seven people, including 27 static movements and 5 dynamic movements. Each movement is repeated 10 times, totaling about 112 min. In particular, we obtained data containing occlusions. Experimental results show that our system can fuse multimodal information and perform better than using only skeletal information, with the accuracy increasing from 68.34% to 84.13%.

GLMDriveNet: Global–local Multimodal Fusion Driving Behavior Classification Network

Driving behavior classification plays an important role in many fields, such as Advanced Driving Assistance System (ADAS), traffic safety, and energy saving. In this paper, we propose a Global–local Multimodal Fusion Driving Behavior Classification Network (GLMDriveNet) which classifies driver behaviors into normal driving, aggressive driving, and drowsy driving. First of all, we design a Global–local Interaction Channel Attention Module (GLI-CAM) to extract effective features in both the roadside image and the spectrogram generated from the current prediction time and its previous four seconds of vehicle speeds. Furthermore, a learnable positional embedding is introduced to fuse the global and local information of the channels for better screening of the extracted features. Secondly, we propose a Multi-scale Feature Representation Fusion Module (MS-FRFM) to associate the high-scale and low-scale information of images and spectrograms and assign different importances for different modal information, making the network more inclined to useful modal information. Our model is evaluated on a public dataset UAH-DriveSet and achieves the best performance (98.4% F1-score on all roads, 97.4% F1-score on the motorway road, and 99.8% F1-score on the secondary road) compared to other state-of-the-art methods. Our model has a very fast speed (142 FPS) and strong generalization which has been verified through extensive experiments on multiple datasets. The code is available on https://github.com/liuwenzhuo1/GLMDrivenet.

Biometric Analysis on Smart Textile Garment in Real Life Scenario

The rapid proliferation of wearable applications and technologies capable of acquiring biomedical signals has prompted the incorporation of biomedical signals, such as the electrocardiogram (ECG), for biometric purposes in wearable platforms. Most ECG biometric research utilises medical-grade sensors in clinical settings, which is unrealistic for wearable ECG-based biometric applications in the real world. Therefore, this research aims to examine the ECG biometric on smart textile garments in real life, collected from commercially available wearable Hexoskin Proshirt and HeartIn Fit shirts. ECG data were obtained from 22 participants who took part in this study. The raw ECG signal is initially pre-processed using noise-removal Butterworth filters in the time domain, followed by an effective QRS segmented feature extraction technique. Finally, around 2076 datasets were created for training and validation, while the remaining 501 datasets were employed to test the suggested recognition approach with 29 Machine Learning Classifiers. Subsequently, Quadratic SVM has the highest accuracy at 96.8% for ECG biometrics, followed by Narrow Neural Network with 95.8% and Wide Neural Network with 95.4%. Further improvement to the QSVM parameter improved the accuracy to 97.4% with an error rate of 2.6%, followed by a sensitivity of 97.4% with a precision of 97.7% and a false rejection rate of 2.6%. Thus, the results of this study further validate the feasibility of applying ECG biometrics for recognition in real-life scenarios utilising a smart textile shirt with different configurations and brand is possible.

Convolutional transformer network for fine-grained action recognition

Fine-grained action recognition is one of the critical problems in video processing, which aims to recognize similar actions of subtle interactions between humans and objects. Inspired by the remarkable performance of the Transformer in natural language processing, Transformer has been applied to the fine-grained action recognition task. However, Transformer needs abundant training data and extra supervision to achieve comparable results with convolutional neural networks (CNNs). To address these issues, we propose a Convolutional Transformer Network (CTN), which integrates the merits of CNN (e.g., sharing weights, capturing low-level features in videos and locality) and the benefits of Transformer (e.g., dynamic attention and learning long-range dependencies). In this paper, we propose two modifications to the original Transformer: (i) We propose a video-to-tokens module that can extract tokens from extracted spatial-temporal features in videos by 3D convolutions instead of the direct token embedding from raw input video clips; (ii) We completely replace the linear mapping in multi-head self-attention layer with depth-wise convolutional mapping, which applies a depth-wise separable convolution operation on embedded token maps. With these two modifications, our approach can extract effective spatial-temporal features from videos and process the long sequences of tokens encountered in videos. Experimental results demonstrate that our proposed CTN can achieve state-of-the-art accuracy on two fine-grained action recognition datasets (i.e., Epic-Kitchens and Diving 48) with a small computational increase.

Lag‐related noise shrinkage stacked LSTM network for short‐term traffic flow forecasting

AbstractFor the transport networks only equipped with sparse or isolated detectors, short‐term traffic flow forecasting faces the following problems: (1) there are only temporal information and no spatial information; (2) the noises in the traffic flow significantly affect the forecasting performance. In this paper, a lag‐related noise shrinkage stacked long short‐term memory (LSTM) network is proposed for the traffic flow forecasting task only related to temporal information. To extract effective temporal features, the optimal time lags are selected in the traffic flow and converted into lag‐related multi‐dimensional data. Then, a discrete wavelet threshold denoising shrinkage algorithm is designed to filter the noises to construct a more reliable training set. A multi‐level stacked LSTM network is employed to learn the features of the training set to map the past traffic flow to the future flow. Four benchmark datasets are to evaluate the forecasting performance by extensive experiments. The comparison with the state‐of‐the‐art models demonstrates an average improvement of 7.28% in MAPE and 6.02% in RMSE. In addition, the proposed method has been applied in the Guilin Travel Network Bus Intelligent Dispatching System. It improves the utilization of the vehicles and reduces operating costs.

Multilinear subspace learning for Person Re-Identification based fusion of high order tensor features

Video surveillance image analysis and processing is an important field in computer vision and among the challenging tasks for Person Re-Identification (PRe-ID). The latter aims at finding a target person who has already been identified and appeared on a camera network using a powerful description of their pedestrian images. The success of recent research on person PRe-ID is largely backed into the effective features extraction and representation with a powerful learning of these features to correctly discriminate pedestrian images. To this end, two powerful features, Convolutional Neural Network (CNN) and Local Maximal Occurrence (LOMO) are modeled on a multidimensional data in the proposed method, High-Dimensional Feature Fusion (HDFF). Specifically, a new tensor fusion scheme is introduced to take advantage and combine two types of features in the same tensor data even if its dimensions are not the same. To improve the accuracy, we use Tensor Cross-View Quadratic Analysis (TXQDA) to perform multilinear subspace learning followed by the Cosine similarity for matching. TXQDA efficiently ensures the learning ability and reduces the high dimensionality resulting from high-order tensor data. The effectiveness of the proposed method is verified through experiments on three challenging widely-used PRe-ID datasets namely, VIPeR, GRID, and PRID450S. Extensive experiments show that the proposed method performs very well when compared with recent state-of-the-art methods.

BEMF-Net: Semantic Segmentation of Large-Scale Point Clouds via Bilateral Neighbor Enhancement and Multi-Scale Fusion

The semantic segmentation of point clouds is a crucial undertaking in 3D reconstruction and holds great importance. However, achieving precise semantic segmentation represents a significant hurdle. In this paper, we present BEMF-Net, an efficient method for large-scale environments. It starts with an effective feature extraction method. Unlike images, 3D data comprise not only geometric relations but also texture information. To accurately depict the scene, it is crucial to take into account the impacts of texture and geometry on the task, and incorporate modifications to improve feature description. Additionally, we present a multi-scale feature fusion technique that effectively promotes the interaction between features at different resolutions. The approach mitigates the problem of the smoothing of detailed information caused by downsampling mechanisms, while ensuring the integrity of features across different layers, allowing a more comprehensive representation of the point cloud. We confirmed the effectiveness of this method by testing it on benchmark datasets such as S3DIS, SensatUrban, and Toronto3D.

Imbalanced learning for wind turbine blade icing detection via spatio-temporal attention model with a self-adaptive weight loss function

Accurate blade icing detection of wind turbines is of great significance to avoid secondary damages or accidents. Recently, data-driven methods have attracted increasing attention due to the availability of supervisory control and data acquisition (SCADA) data with their low-cost and high volume. However, SCADA data usually present complex characteristics of high dimensionality and strong spatio-temporal correlations among different sensor variables and severe class imbalance. It is challenging to extract effective features for accurate and reliable icing detection from imbalanced SCADA data. To address these challenges, we propose an imbalanced learning method for blade icing detection. First, we propose a spatio-temporal attention model to extract and assign weights to temporal-spatial features based on their contributions to blade icing detection. Then, to address the class imbalance issue, we design a self-adaptive weight (SAW) loss function to divide SCADA data into multiple batches and then classify them during model training. Unlike conventional loss functions, our SAW loss can adaptively assign weights to each data category according to their numbers in the divided batches. The effectiveness of the proposed method is verified on real SCADA datasets. The results demonstrate our designed spatio-temporal attention and SAW loss achieved better classification results and superior self-adaptability.

An effective feature extraction method for olive peacock eye leaf disease classification

An effective feature extraction method for olive peacock eye leaf disease classification

XCNN-SC: Explainable CNN for SARS-CoV-2 variants classification and mutation detection

The COVID-19 pandemic spread rapidly all over the world in 2019, causing the deaths of millions of people. One of the main challenges in controlling the pandemic was the high rate of virus mutation, which evaded the immune system and reduced the vaccine's effectiveness. Due to the differences in symptoms, transmission and mortality rate, and effective prevention strategies of each variant, identifying and classifying different variants is a great necessity. In the present study, 82802 whole genomes of Alpha, Beta, Delta, Eta, Epsilon, Lota, and Omicron variants of SARS-CoV-2 were obtained from the NCBI database. The label encoding method was applied to convert the genomic sequences to numeric sequences. Then, a single-layer 1D-CNN model was used to classify seven variants of SARS-CoV-2 and achieved 99.78% accuracy, 97.98% precision, 97.66% sensitivity, 99.95% specificity, and 98.68% F1-score. The max pool layer feature maps of this network were investigated to discover the sequence discriminative regions in SARS-CoV-2 variants and their effect on biological functional differences. The feature map examination led to the detection of three mutations; one of them was a missense mutation, which has been reported previously. The rest of the detected mutations were silent mutations. In conclusion, the achieved results indicated that CNN models can extract effective features to discriminate several SARS-CoV-2 variants. Also, investigation of CNN feature maps led to an explainable CNN revealing the learned knowledge of the network from the training database. This knowledge helped us to detect mutations and their functional effects in different variants.

Image inpainting network based on multi‐level attention mechanism

AbstractImage inpainting networks based on deep learning techniques have been widely used in many important fields. However, most inpainting networks fail to generate desirable repaired images. This may be due to their failure to extract effective features and accurately assign high weights to the undamaged regions. To alleviate these problems, an image inpainting network based on gated convolution and multi‐level attention mechanism (IIN‐GCMAM) is proposed in this paper. This network follows encoder–decoder architecture, consisting of the gated convolution encoder (GC‐encoder) and the multi‐level attention mechanism decoder (MAM‐decoder). The GC‐encoder weighs the extracted features with gated convolutions, which reduces the interference caused by the damaged regions. The multi‐level attention mechanism employed in the MAM‐decoder uses multi‐scale feature maps spatially and channel‐wise to improve the consistency in global structure and the fineness of repaired results. Extensive experiments are conducted on the common datasets, Paris StreetView and CelebA. Experimental results indicate that the proposed IIN‐GCMAM can achieve a good performance on the common evaluation metrics and visual effects. It can achieve 0.0408, 0.720, and 22.27 in MAE, SSIM, and PSNR at the mask ratio of 50%–60%, respectively.