Disordered Multi-view Registration Method Based on the Soft Trimmed Deep Network

Compared with the pair-wise registration of point clouds,multi-view point cloud registration is much less studied.In this dissertation,a disordered multi-view point cloud registration method based on the soft trimmed deep network is proposed.In this method,firstly,the expression ability of feature extraction module is improved and the registration accuracy is increased by enhancing feature extraction network with the point pair feature.Secondly,neighborhood and angle similarities are used to measure the consistency of candidate points to surrounding neighborhoods.By combining distance consistency and high dimensional feature consistency,our network introduces the confidence estimation module of registration,so the point cloud trimmed problem can be converted to candidate for the degree of confidence estimation problem,achieving the pair-wise registration of partially overlapping point clouds.Thirdly,the results from pair-wise registration are fed into the model fusion to achieve the rough registration of multi-view point clouds.Finally,the hierarchical clustering is used to iteratively optimize the clustering center model by gradually increasing the number of clustering categories and performing clustering and registration alternately.This method achieves rough point cloud registration quickly in the early stage,improves the accuracy of multi-view point cloud registration in the later stage,and makes full use of global information to achieve robust and accurate multi-view registration without initial value.

Face recognition using both visible light image and near-infrared image and a deep network

In recent years, deep networks has achieved outstanding performance in computer vision, especially in the field of face recognition. In terms of the performance for a face recognition model based on deep network, there are two main closely related factors： 1） the structure of the deep neural network, and 2） the number and quality of training data. In real applications, illumination change is one of the most important factors that significantly affect the performance of face recognition algorithms. As for deep network models, only if there is sufficient training data that has various illumination intensity could they achieve expected performance. However, such kind of training data is hard to collect in the real world. In this paper, focusing on the illumination change challenge, we propose a deep network model which takes both visible light image and near-infrared image into account to perform face recognition. Near- infrared image, as we know, is much less sensitive to illuminations. Visible light face image contains abundant texture information which is very useful for face recognition. Thus, we design an adaptive score fusion strategy which hardly has information loss and the nearest neighbor algorithm to conduct the final classification. The experimental results demonstrate that the model is very effective in realworld scenarios and perform much better in terms of illumination change than other state-of-the-art models.

DNEF:A New Ensemble Framework Based on Deep Network Structure

Deep neural networks have achieved tremendous success in various fields,and the structure of these networks is a key factor in their success.In this paper,we focus on the research of ensemble learning based on deep network structure and propose a new deep network ensemble framework(DNEF).Unlike other ensemble learning models,DNEF is an ensemble learning architecture of network structures,with serial iteration between the hidden layers,while base classifiers are trained in parallel within these hidden layers.Specifically,DNEF uses randomly sampled data as input and implements serial iteration based on the weighting strategy between hidden layers.In the hidden layers,each node represents a base classifier,and multiple nodes generate training data for the next hidden layer according to the transfer strategy.The DNEF operates based on two strategies:(1)The weighting strategy calculates the training instance weights of the nodes according to their weaknesses in the previous layer.(2)The transfer strategy adaptively selects each node’s instances with weights as transfer instances and transfer weights,which are combined with the training data of nodes as input for the next hidden layer.These two strategies improve the accuracy and generalization of DNEF.This research integrates the ensemble of all nodes as the final output of DNEF.The experimental results reveal that the DNEF framework surpasses the traditional ensemble models and functions with high accuracy and innovative deep ensemble methods.

Action Recognition in Surveillance Videos with Combined Deep Network Models

Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, most existing deep learning based recognition frameworks are not optimized for action in the surveillance videos. In this paper, we propose a novel method to deal with the recognition of different types of actions in outdoor surveillance videos. The proposed method first introduces motion compensation to improve the detection of human target. Then, it uses three different types of deep models with single and sequenced images as inputs for the recognition of different types of actions. Finally, predictions from different models are fused with a linear model. Experimental results show that the proposed method works well on the real surveillance videos.

Detecting and Classifying Darknet Traffic Using Deep Network Chains

The anonymity of the darknet makes it attractive to secure communication lines from censorship.The analysis,monitoring,and categorization of Internet network traffic are essential for detecting darknet traffic that can generate a comprehensive characterization of dangerous users and assist in tracing malicious activities and reducing cybercrime.Furthermore,classifying darknet traffic is essential for real-time applications such as the timely monitoring of malware before attacks occur.This paper presents a two-stage deep network chain for detecting and classifying darknet traffic.In the first stage,anonymized darknet traffic,including VPN and Tor traffic related to hidden services provided by darknets,is detected.In the second stage,traffic related to VPNs and Tor services is classified based on their respective applications.The methodology of this paper was verified on a benchmark dataset containing VPN and Tor traffic.It achieved an accuracy of 96.8%and 94.4%in the detection and classification stages,respectively.Optimization and parameter tuning were performed in both stages to achieve more accurate results,enabling practitioners to combat alleged malicious activities and further detect such activities after outbreaks.In the classification stage,it was observed that the misclassifications were due to the audio and video streaming commonly used in shared real-time protocols.However,in cases where it is desired to distinguish between such activities accurately,the presented deep chain classifier can accommodate additional classifiers.Furthermore,additional classifiers could be added to the chain to categorize specific activities of interest further.

A novel interpretable multilevel wavelet decomposition deep network for actual heartbeat classification

Arrhythmias may lead to sudden cardiac death if not detected and treated in time.A supraventricular premature beat(SPB)and premature ventricular contraction(PVC)are important categories of arrhythmia disease.Recently,deep learning methods have been applied to the PVC/SPB heartbeats detection.However,most researchers have focused on time-domain information of the electrocardiogram and there has been a lack of exploration of the interpretability of the model.In this study,we design an interpretable and accurate PVC/SPB recognition algorithm,called the interpretable multilevel wavelet decomposition deep network(IMWDDN).Wavelet decomposition is introduced into the deep network and the squeeze and excitation(SE)-Residual block is designed for extracting time-domain and frequency-domain features.Additionally,inspired by the idea of residual learning,we construct a novel loss function for the constant updating of the multilevel wavelet decomposition parameters.Finally,the IMWDDN is evaluated on the Third China Physiological Signal Challenge Dataset and the MIT-BIH Arrhythmia database.The comparison results show IMWDDN has better detection performance with 98.51%accuracy and a 93.75%F1-macro on average,and its areas of concern are similar to those of an expert diagnosis to a certain extent.Generally,the IMWDDN has good application value in the clinical screening of PVC/SPB heartbeats.

Hybrid model for BOF oxygen blowing time prediction based on oxygen balance mechanism and deep neural network

The amount of oxygen blown into the converter is one of the key parameters for the control of the converter blowing process,which directly affects the tap-to-tap time of converter. In this study, a hybrid model based on oxygen balance mechanism (OBM) and deep neural network (DNN) was established for predicting oxygen blowing time in converter. A three-step method was utilized in the hybrid model. First, the oxygen consumption volume was predicted by the OBM model and DNN model, respectively. Second, a more accurate oxygen consumption volume was obtained by integrating the OBM model and DNN model. Finally, the converter oxygen blowing time was calculated according to the oxygen consumption volume and the oxygen supply intensity of each heat. The proposed hybrid model was verified using the actual data collected from an integrated steel plant in China, and compared with multiple linear regression model, OBM model, and neural network model including extreme learning machine, back propagation neural network, and DNN. The test results indicate that the hybrid model with a network structure of 3 hidden layer layers, 32-16-8 neurons per hidden layer, and 0.1 learning rate has the best prediction accuracy and stronger generalization ability compared with other models. The predicted hit ratio of oxygen consumption volume within the error±300 m^(3)is 96.67%;determination coefficient (R^(2)) and root mean square error (RMSE) are0.6984 and 150.03 m^(3), respectively. The oxygen blow time prediction hit ratio within the error±0.6 min is 89.50%;R2and RMSE are0.9486 and 0.3592 min, respectively. As a result, the proposed model can effectively predict the oxygen consumption volume and oxygen blowing time in the converter.

Fast solution to the free return orbit's reachable domain of the manned lunar mission by deep neural network

It is important to calculate the reachable domain(RD)of the manned lunar mission to evaluate whether a lunar landing site could be reached by the spacecraft. In this paper, the RD of free return orbits is quickly evaluated and calculated via the classification and regression neural networks. An efficient databasegeneration method is developed for obtaining eight types of free return orbits and then the RD is defined by the orbit’s inclination and right ascension of ascending node(RAAN) at the perilune. A classify neural network and a regression network are trained respectively. The former is built for classifying the type of the RD, and the latter is built for calculating the inclination and RAAN of the RD. The simulation results show that two neural networks are well trained. The classification model has an accuracy of more than 99% and the mean square error of the regression model is less than 0.01°on the test set. Moreover, a serial strategy is proposed to combine the two surrogate models and a recognition tool is built to evaluate whether a lunar site could be reached. The proposed deep learning method shows the superiority in computation efficiency compared with the traditional double two-body model.

Self-potential inversion based on Attention U-Net deep learning network

Landfill leaks pose a serious threat to environmental health,risking the contamination of both groundwater and soil resources.Accurate investigation of these sites is essential for implementing effective prevention and control measures.The self-potential(SP)stands out for its sensitivity to contamination plumes,offering a solution for monitoring and detecting the movement and seepage of subsurface pollutants.However,traditional SP inversion techniques heavily rely on precise subsurface resistivity information.In this study,we propose the Attention U-Net deep learning network for rapid SP inversion.By incorporating an attention mechanism,this algorithm effectively learns the relationship between array-style SP data and the location and extent of subsurface contaminated sources.We designed a synthetic landfill model with a heterogeneous resistivity structure to assess the performance of Attention U-Net deep learning network.Additionally,we conducted further validation using a laboratory model to assess its practical applicability.The results demonstrate that the algorithm is not solely dependent on resistivity information,enabling effective locating of the source distribution,even in models with intricate subsurface structures.Our work provides a promising tool for SP data processing,enhancing the applicability of this method in the field of near-subsurface environmental monitoring.

Network Security Enhanced with Deep Neural Network-Based Intrusion Detection System

This study describes improving network security by implementing and assessing an intrusion detection system(IDS)based on deep neural networks(DNNs).The paper investigates contemporary technical ways for enhancing intrusion detection performance,given the vital relevance of safeguarding computer networks against harmful activity.The DNN-based IDS is trained and validated by the model using the NSL-KDD dataset,a popular benchmark for IDS research.The model performs well in both the training and validation stages,with 91.30%training accuracy and 94.38%validation accuracy.Thus,the model shows good learning and generalization capabilities with minor losses of 0.22 in training and 0.1553 in validation.Furthermore,for both macro and micro averages across class 0(normal)and class 1(anomalous)data,the study evaluates the model using a variety of assessment measures,such as accuracy scores,precision,recall,and F1 scores.The macro-average recall is 0.9422,the macro-average precision is 0.9482,and the accuracy scores are 0.942.Furthermore,macro-averaged F1 scores of 0.9245 for class 1 and 0.9434 for class 0 demonstrate the model’s ability to precisely identify anomalies precisely.The research also highlights how real-time threat monitoring and enhanced resistance against new online attacks may be achieved byDNN-based intrusion detection systems,which can significantly improve network security.The study underscores the critical function ofDNN-based IDS in contemporary cybersecurity procedures by setting the foundation for further developments in this field.Upcoming research aims to enhance intrusion detection systems by examining cooperative learning techniques and integrating up-to-date threat knowledge.

Diffraction deep neural network-based classification for vector vortex beams

The vector vortex beam(VVB)has attracted significant attention due to its intrinsic diversity of information and has found great applications in both classical and quantum communications.However,a VVB is unavoidably affected by atmospheric turbulence(AT)when it propagates through the free-space optical communication environment,which results in detection errors at the receiver.In this paper,we propose a VVB classification scheme to detect VVBs with continuously changing polarization states under AT,where a diffractive deep neural network(DDNN)is designed and trained to classify the intensity distribution of the input distorted VVBs,and the horizontal direction of polarization of the input distorted beam is adopted as the feature for the classification through the DDNN.The numerical simulations and experimental results demonstrate that the proposed scheme has high accuracy in classification tasks.The energy distribution percentage remains above 95%from weak to medium AT,and the classification accuracy can remain above 95%for various strengths of turbulence.It has a faster convergence and better accuracy than that based on a convolutional neural network.

Geometric prior guided hybrid deep neural network for facial beauty analysis

Facial beauty analysis is an important topic in human society.It may be used as a guidance for face beautification applications such as cosmetic surgery.Deep neural networks(DNNs)have recently been adopted for facial beauty analysis and have achieved remarkable performance.However,most existing DNN-based models regard facial beauty analysis as a normal classification task.They ignore important prior knowledge in traditional machine learning models which illustrate the significant contribution of the geometric features in facial beauty analysis.To be specific,landmarks of the whole face and facial organs are introduced to extract geometric features to make the decision.Inspired by this,we introduce a novel dual-branch network for facial beauty analysis:one branch takes the Swin Transformer as the backbone to model the full face and global patterns,and another branch focuses on the masked facial organs with the residual network to model the local patterns of certain facial parts.Additionally,the designed multi-scale feature fusion module can further facilitate our network to learn complementary semantic information between the two branches.In model optimisation,we propose a hybrid loss function,where especially geometric regulation is introduced by regressing the facial landmarks and it can force the extracted features to convey facial geometric features.Experiments performed on the SCUT-FBP5500 dataset and the SCUT-FBP dataset demonstrate that our model outperforms the state-of-the-art convolutional neural networks models,which proves the effectiveness of the proposed geometric regularisation and dual-branch structure with the hybrid network.To the best of our knowledge,this is the first study to introduce a Vision Transformer into the facial beauty analysis task.

Radar Signal Intra-Pulse Modulation Recognition Based on Deep Residual Network

In view of low recognition rate of complex radar intra-pulse modulation signal type by traditional methods under low signal-to-noise ratio(SNR),the paper proposes an automatic recog-nition method of complex radar intra-pulse modulation signal type based on deep residual network.The basic principle of the recognition method is to obtain the transformation relationship between the time and frequency of complex radar intra-pulse modulation signal through short-time Fourier transform(STFT),and then design an appropriate deep residual network to extract the features of the time-frequency map and complete a variety of complex intra-pulse modulation signal type recognition.In addition,in order to improve the generalization ability of the proposed method,label smoothing and L2 regularization are introduced.The simulation results show that the proposed method has a recognition accuracy of more than 95%for complex radar intra-pulse modulation sig-nal types under low SNR(2 dB).

Probabilistic seismic inversion based on physics-guided deep mixture density network

Deterministic inversion based on deep learning has been widely utilized in model parameters estimation.Constrained by logging data,seismic data,wavelet and modeling operator,deterministic inversion based on deep learning can establish nonlinear relationships between seismic data and model parameters.However,seismic data lacks low-frequency and contains noise,which increases the non-uniqueness of the solutions.The conventional inversion method based on deep learning can only establish the deterministic relationship between seismic data and parameters,and cannot quantify the uncertainty of inversion.In order to quickly quantify the uncertainty,a physics-guided deep mixture density network(PG-DMDN)is established by combining the mixture density network(MDN)with the deep neural network(DNN).Compared with Bayesian neural network(BNN)and network dropout,PG-DMDN has lower computing cost and shorter training time.A low-frequency model is introduced in the training process of the network to help the network learn the nonlinear relationship between narrowband seismic data and low-frequency impedance.In addition,the block constraints are added to the PG-DMDN framework to improve the horizontal continuity of the inversion results.To illustrate the benefits of proposed method,the PG-DMDN is compared with existing semi-supervised inversion method.Four synthetic data examples of Marmousi II model are utilized to quantify the influence of forward modeling part,low-frequency model,noise and the pseudo-wells number on inversion results,and prove the feasibility and stability of the proposed method.In addition,the robustness and generality of the proposed method are verified by the field seismic data.

Developing a novel big dataset and a deep neural network to predict the bearing capacity of a ring footing

The accurate prediction of the bearing capacity of ring footings,which is crucial for civil engineering projects,has historically posed significant challenges.Previous research in this area has been constrained by considering only a limited number of parameters or utilizing relatively small datasets.To overcome these limitations,a comprehensive finite element limit analysis(FELA)was conducted to predict the bearing capacity of ring footings.The study considered a range of effective parameters,including clay undrained shear strength,heterogeneity factor of clay,soil friction angle of the sand layer,radius ratio of the ring footing,sand layer thickness,and the interface between the ring footing and the soil.An extensive dataset comprising 80,000 samples was assembled,exceeding the limitations of previous research.The availability of this dataset enabled more robust and statistically significant analyses and predictions of ring footing bearing capacity.In light of the time-intensive nature of gathering a substantial dataset,a customized deep neural network(DNN)was developed specifically to predict the bearing capacity of the dataset rapidly.Both computational and comparative results indicate that the proposed DNN(i.e.DNN-4)can accurately predict the bearing capacity of a soil with an R2 value greater than 0.99 and a mean squared error(MSE)below 0.009 in a fraction of 1 s,reflecting the effectiveness and efficiency of the proposed method.

Improved Double Deep Q Network Algorithm Based on Average Q-Value Estimation and Reward Redistribution for Robot Path Planning

By integrating deep neural networks with reinforcement learning,the Double Deep Q Network(DDQN)algorithm overcomes the limitations of Q-learning in handling continuous spaces and is widely applied in the path planning of mobile robots.However,the traditional DDQN algorithm suffers from sparse rewards and inefficient utilization of high-quality data.Targeting those problems,an improved DDQN algorithm based on average Q-value estimation and reward redistribution was proposed.First,to enhance the precision of the target Q-value,the average of multiple previously learned Q-values from the target Q network is used to replace the single Q-value from the current target Q network.Next,a reward redistribution mechanism is designed to overcome the sparse reward problem by adjusting the final reward of each action using the round reward from trajectory information.Additionally,a reward-prioritized experience selection method is introduced,which ranks experience samples according to reward values to ensure frequent utilization of high-quality data.Finally,simulation experiments are conducted to verify the effectiveness of the proposed algorithm in fixed-position scenario and random environments.The experimental results show that compared to the traditional DDQN algorithm,the proposed algorithm achieves shorter average running time,higher average return and fewer average steps.The performance of the proposed algorithm is improved by 11.43%in the fixed scenario and 8.33%in random environments.It not only plans economic and safe paths but also significantly improves efficiency and generalization in path planning,making it suitable for widespread application in autonomous navigation and industrial automation.

Data-Driven Modeling for Wind Turbine Blade Loads Based on Deep Neural Network

Blades are essential components of wind turbines.Reducing their fatigue loads during operation helps to extend their lifespan,but it is difficult to quickly and accurately calculate the fatigue loads of blades.To solve this problem,this paper innovatively designs a data-driven blade load modeling method based on a deep learning framework through mechanism analysis,feature selection,and model construction.In the mechanism analysis part,the generation mechanism of blade loads and the load theoretical calculationmethod based on material damage theory are analyzed,and four measurable operating state parameters related to blade loads are screened;in the feature extraction part,15 characteristic indicators of each screened parameter are extracted in the time and frequency domain,and feature selection is completed through correlation analysis with blade loads to determine the input parameters of data-driven modeling;in the model construction part,a deep neural network based on feedforward and feedback propagation is designed to construct the nonlinear coupling relationship between the unit operating parameter characteristics and blade loads.The results show that the proposed method mines the wind turbine operating state characteristics highly correlated with the blade load,such as the standard deviation of wind speed.The model built using these characteristics has reasonable calculation and fitting capabilities for the blade load and shows a better fitting level for untrained out-of-sample data than the traditional scheme.Based on the mean absolute percentage error calculation,the modeling accuracy of the two blade loads can reach more than 90%and 80%,respectively,providing a good foundation for the subsequent optimization control to suppress the blade load.

Nonparametric Statistical Feature Scaling Based Quadratic Regressive Convolution Deep Neural Network for Software Fault Prediction

The development of defect prediction plays a significant role in improving software quality. Such predictions are used to identify defective modules before the testing and to minimize the time and cost. The software with defects negatively impacts operational costs and finally affects customer satisfaction. Numerous approaches exist to predict software defects. However, the timely and accurate software bugs are the major challenging issues. To improve the timely and accurate software defect prediction, a novel technique called Nonparametric Statistical feature scaled QuAdratic regressive convolution Deep nEural Network (SQADEN) is introduced. The proposed SQADEN technique mainly includes two major processes namely metric or feature selection and classification. First, the SQADEN uses the nonparametric statistical Torgerson–Gower scaling technique for identifying the relevant software metrics by measuring the similarity using the dice coefficient. The feature selection process is used to minimize the time complexity of software fault prediction. With the selected metrics, software fault perdition with the help of the Quadratic Censored regressive convolution deep neural network-based classification. The deep learning classifier analyzes the training and testing samples using the contingency correlation coefficient. The softstep activation function is used to provide the final fault prediction results. To minimize the error, the Nelder–Mead method is applied to solve non-linear least-squares problems. Finally, accurate classification results with a minimum error are obtained at the output layer. Experimental evaluation is carried out with different quantitative metrics such as accuracy, precision, recall, F-measure, and time complexity. The analyzed results demonstrate the superior performance of our proposed SQADEN technique with maximum accuracy, sensitivity and specificity by 3%, 3%, 2% and 3% and minimum time and space by 13% and 15% when compared with the two state-of-the-art methods.

Enhancing Human Action Recognition with Adaptive Hybrid Deep Attentive Networks and Archerfish Optimization

In recent years,wearable devices-based Human Activity Recognition(HAR)models have received significant attention.Previously developed HAR models use hand-crafted features to recognize human activities,leading to the extraction of basic features.The images captured by wearable sensors contain advanced features,allowing them to be analyzed by deep learning algorithms to enhance the detection and recognition of human actions.Poor lighting and limited sensor capabilities can impact data quality,making the recognition of human actions a challenging task.The unimodal-based HAR approaches are not suitable in a real-time environment.Therefore,an updated HAR model is developed using multiple types of data and an advanced deep-learning approach.Firstly,the required signals and sensor data are accumulated from the standard databases.From these signals,the wave features are retrieved.Then the extracted wave features and sensor data are given as the input to recognize the human activity.An Adaptive Hybrid Deep Attentive Network(AHDAN)is developed by incorporating a“1D Convolutional Neural Network(1DCNN)”with a“Gated Recurrent Unit(GRU)”for the human activity recognition process.Additionally,the Enhanced Archerfish Hunting Optimizer(EAHO)is suggested to fine-tune the network parameters for enhancing the recognition process.An experimental evaluation is performed on various deep learning networks and heuristic algorithms to confirm the effectiveness of the proposed HAR model.The EAHO-based HAR model outperforms traditional deep learning networks with an accuracy of 95.36,95.25 for recall,95.48 for specificity,and 95.47 for precision,respectively.The result proved that the developed model is effective in recognizing human action by taking less time.Additionally,it reduces the computation complexity and overfitting issue through using an optimization approach.

The Short-Term Prediction ofWind Power Based on the Convolutional Graph Attention Deep Neural Network

The fluctuation of wind power affects the operating safety and power consumption of the electric power grid and restricts the grid connection of wind power on a large scale.Therefore,wind power forecasting plays a key role in improving the safety and economic benefits of the power grid.This paper proposes a wind power predicting method based on a convolutional graph attention deep neural network with multi-wind farm data.Based on the graph attention network and attention mechanism,the method extracts spatial-temporal characteristics from the data of multiple wind farms.Then,combined with a deep neural network,a convolutional graph attention deep neural network model is constructed.Finally,the model is trained with the quantile regression loss function to achieve the wind power deterministic and probabilistic prediction based on multi-wind farm spatial-temporal data.A wind power dataset in the U.S.is taken as an example to demonstrate the efficacy of the proposed model.Compared with the selected baseline methods,the proposed model achieves the best prediction performance.The point prediction errors(i.e.,root mean square error(RMSE)and normalized mean absolute percentage error(NMAPE))are 0.304 MW and 1.177%,respectively.And the comprehensive performance of probabilistic prediction(i.e.,con-tinuously ranked probability score(CRPS))is 0.580.Thus,the significance of multi-wind farm data and spatial-temporal feature extraction module is self-evident.