Search

Generative adversarial networks (GANs) have recently achieved impressive results in facial age synthesis. However, these methods usually select an autoencoder-style generator. And the bottleneck layer in the encoder-decoder generally gives rise to blurry and low-quality generation. To address this limitation, we propose a novel attention-aware conditional generative adversarial network (ACGAN). First, we utilize two different attention mechanisms to improve generation quality. On one hand, we integrate channel attention modules into the generator to enhance the discriminative representation power. On the other hand, we introduce a position attention mask to well-process images captured with various backgrounds and illuminations. Second, we deploy a local discriminator to enhance the central face region with informative details. Third, we adopt three types of losses to achieve accurate age generation and preserve personalized features: 1) The adversarial loss aims to synthesize photo-realistic faces with expected aging effects; 2) The identity loss intends to keep identity information unchanged; 3) The attention loss tries to improve the accuracy of attention mask regression. To assess the effectiveness of the proposed method, we conduct extensive experiments on several public aging databases. Experimental results on MORPH, CACD, and FG-NET show the effectiveness of the proposed framework.

This article aims to tackle the problem of group activity recognition in the multiple-person scene. To model the group activity with multiple persons, most long short-term memory (LSTM)-based methods first learn the person-level action representations by several LSTMs and then integrate all the person-level action representations into the following LSTM to learn the group-level activity representation. This type of solution is a two-stage strategy, which neglects the "host-parasite" relationship between the group-level activity ("host") and person-level actions ("parasite") in spatiotemporal space. To this end, we propose a novel graph LSTM-in-LSTM (GLIL) for group activity recognition by modeling the person-level actions and the group-level activity simultaneously. GLIL is a "host-parasite" architecture, which can be seen as several person LSTMs (P-LSTMs) in the local view or a graph LSTM (G-LSTM) in the global view. Specifically, P-LSTMs model the person-level actions based on the interactions among persons. Meanwhile, G-LSTM models the group-level activity, where the person-level motion information in multiple P-LSTMs is selectively integrated and stored into G-LSTM based on their contributions to the inference of the group activity class. Furthermore, to use the person-level temporal features instead of the person-level static features as the input of GLIL, we introduce a residual LSTM with the residual connection to learn the person-level residual features, consisting of temporal features and static features. Experimental results on two public data sets illustrate the effectiveness of the proposed GLIL compared with state-of-the-art methods.

This work aims to freely translate an input face to an aging face with robust identity preservation, satisfying aging effect and authentic visual appearance. Witnessing the success of GAN in image synthesis, researchers employ GAN to address the problem of face aging synthesis. However, most GAN-based methods hold that the aging changing of all facial regions is equal, which ignores the fact that different facial regions have distinct aging speeds and aging patterns. To this end, we propose a novel Conditioned-Attention Normalization GAN (CAN-GAN) for age synthesis by leveraging the aging difference between two age groups to capture facial aging regions with different attention factors. In particular, a new Conditioned-Attention Normalization (CAN) layer is designed to enhance the aging-relevant information of face, while smoothing the aging-irrelevant information of face by attention map. Since different facial attributes contribute to the discrimination of age groups with divers degrees, we further present a Contribution-Aware Age Classifier (CAAC) that finely measures the importance of face vector’s elements in terms of the age classification. Qualitative and quantitative experiments on several commonly-used datasets show the advance of CAN-GAN compared with the other competitive methods.

In feature learning tasks, one of the most enormous challenges is to generate an efficient discriminative subspace. In this paper, we propose a novel subspace learning method, named recursive discriminative subspace learning with an ${\ell }_{{1}}$ -norm distance constraint (RDSL). RDSL can robustly extract features from the contaminated images and learn a discriminative subspace. With the use of an inequation-based ${\ell }_{{1}}$ -norm distance metric constraint, the minimized ${\ell }_{{1}}$ -norm distance metric objective function with slack variables induces samples in the same class to cluster as close as possible, meanwhile samples from different classes can be separated from each other as far as possible. By utilizing ${\ell }_{{1}}$ -norm items in both the objective function and the constraint, RDSL can well handle the noisy data and outliers. In addition, the large margin formulation makes the proposed method insensitive to initializations. We describe two approaches to solve RDSL with a recursive strategy. Experimental results on six benchmark datasets, including the original data and the contaminated data, demonstrate that RDSL outperforms the state-of-the-art methods.

Facial image synthesis has been extensively studied, for a long time, in both computer graphics and computer vision. Particularly, the synthesis of face images with varying ages, expressions and poses has received an increasing attention owing to several real-world applications. In this paper, facial age and expression synthesis are addressed. While previous and current research papers on facial age synthesis mostly adopt an age span of 10 years, this paper investigates face aging with a shorter time span. For expression synthesis, given a neutral face, we work on synthesizing faces with varying expression intensities (e.g., from zero to high). Note that both human ages and expression intensities are inherently ordinal. To fully exploit this ordinal nature, we devise ordinal ranking generative adversarial networks (ranking GAN). For each face, a one-hot label is assigned to define its age range/expression intensity. By exploiting the relative order information among age ranges/expression intensities, a binary ranking vector is further computed for each face. In ranking GAN, one-hot labels are used as the condition of the generator for synthesizing faces with target age groups/expression intensities. Moreover, we add a sequence of cost-sensitive ordinal rankers on top of several multi-scale discriminators, with the aim of minimizing age/intensity rank estimation loss when optimizing both the generator and discriminators. In order to evaluate the proposed ranking GAN, extensive experiments are carried out on several public face databases. As demonstrated by the experimental testing, this ranking scheme performs well even when the amount of available labeled training data is limited. The reported experimental results well demonstrate the effectiveness of ranking GAN on synthesizing face aging sequences and faces with varying expression intensities.

The existing research work on facial age synthesis has been mostly focused on long-term aging (e.g., over an age span of 10 years or more). In this paper, we employ generative adversarial networks (GANs) as a tool to investigate age synthesis over different age spans. Compared with long-term aging, short-term age synthesis suffers from the reduced amount of available training data, which can severely hinder the model training. We conduct a series of experiments to validate this. To facilitate short-term age synthesis, we further propose label distribution-guided generative adversarial network (ldGAN), where each sample is associated with an age label distribution (ALD) rather than a single age group. Accordingly, each sample can contribute not only to the learning of its own age group but also to neighbouring groups’ learning. This is useful when addressing short-term aging to cope with the reduced amount of training data. In addition, unlike one-hot encoding which treats age groups as independent from one another, ldGAN can well capture the correlation among different age groups, so that smooth aging sequences can be achieved. The ALD model is integrated into GAN with a two-step process. Firstly, instead of the traditional one-hot encoding, ALD is applied as the condition of the generator. Secondly, we add a sequence of label distribution learners on top of several multi-scale discriminators, with the aim of minimizing the label distribution learning loss when optimizing both the generator and discriminators. Both qualitative and quantitative evaluations are conducted to assess ldGAN’s ability in dealing with two core issues of face aging, i.e., aging effect generation and identity preservation. The obtained experimental results demonstrate the effectiveness of ldGAN in both learning short-term aging patterns and coping with the lack of training data.

Few-shot fine-grained recognition (FS-FGR) aims to recognize novel fine-grained categories with the help of limited available samples. Undoubtedly, this task inherits the main challenges from both few-shot learning and fine-grained recognition. First, the lack of labeled samples makes the learned model easy to overfit. Second, it also suffers from high intra-class variance and low inter-class differences in the datasets. To address this challenging task, we propose a two-stage background suppression and foreground alignment framework, which is composed of a background activation suppression (BAS) module, a foreground object alignment (FOA) module, and a local-to-local (L2L) similarity metric. Specifically, the BAS is introduced to generate a foreground mask for localization to weaken background disturbance and enhance dominative foreground objects. The FOA then reconstructs the feature map of each support sample according to its correction to the query ones, which addresses the problem of misalignment between support-query image pairs. To enable the proposed method to have the ability to capture subtle differences in confused samples, we present a novel L2L similarity metric to further measure the local similarity between a pair of aligned spatial features in the embedding space. What's more, considering that background interference brings poor robustness, we infer the pairwise similarity of feature maps using both the raw image and the refined image. Extensive experiments conducted on multiple popular fine-grained benchmarks demonstrate that our method outperforms the existing state of the art by a large margin. The source codes are available at: https://github.com/CSer-Tang-hao/BSFA-FSFG., Comment: Accepted to IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) 2023

To ensure the photovoltaic (PV) arrays under partial shading condition(PSC) could still output maximum power quickly and efficiently, this work presents a modified hybrid maximum power point tracking (MPPT) method, which applies artificial neural network (ANN) to the modified perturb and observe (MP&O). Instead of using expensive illumination intensity sensors directly, the illumination intensity on each module in the PV array can be obtained indirectly by sampling the specific points of their own cheaper voltage-current sensors. ANN uses indirect illumination intensity to predict the optimal voltage areas of the global maximum power point (GMPP). Based on the areas, MP&O adopts a adaptive step size strategy to obtain GMPP. By modeling and simulation in Matlab/Simulink, it is shown that the tracking time and efficiency of the proposed method in this work can reach 0.026s and 99.87% respectively. Compared with other methods, the method has faster speed, higher efficiency, smaller fluctuation and lower complexity.

As the underlying characteristic distribution differs among different frequencies, existing face inpainting methods on RGB domain can cause some artificial boundaries and blurred details when the masks are large. To reduce this interference, we propose to inpaint face images in wavelet domain with two branches to capture global structure topology and local detailed texture separately. Meanwhile, we adopt the channel relation attention module to learn different weights for different feature channels, with the aim of ensuring the consistency between structure and texture features. In addition, a new wavelet loss is designed to constrain the generated wavelet coefficients closer to the ground truth. Experimental results on CelebA-HQ and Helen datasets demonstrate that our Wavelet Prediction Network outperforms current state-of-the-art face inpainting techniques both qualitatively and quantitatively, especially when handling face images with large masks.

Overlapping community detection, dynamic community identification and community evolution analysis are the three important problems for social network analysis. It is a challenging task to simultaneously address all these three problems with one single method, thus most traditional studies focus on only one or two of them. This paper proposes a novel Dynamic Overlapping Community Evolution Tracking (DOCET) method to solve the three problems simultaneously with one single model, i.e. topology potential field. Specifically, the proposed DOCET method first detects the initial overlapping community structure based on node location analysis in the peak-valley structure of the topology potential field; then it incrementally updates the dynamic community structure based on influence scope analysis in the topology potential field; finally it tracks community evolution events based on the variation of core nodes in the topology potential field. Experiment results on both synthetic and real-world networks show that our proposed method achieves remarkable performance over the existing state-of-the-art methods. It can both accurately partition dynamic overlapping social networks and efficiently track all kinds of community evolution events.

Biometrics is the technique of automatically recognizing individuals based on their biological or behavioral characteristics. Various biometric traits have been introduced and widely investigated, including fingerprint, iris, face, voice, palmprint, gait and so forth. Apart from identity, biometric data may convey various other personal information, covering affect, age, gender, race, accent, handedness, height, weight, etc. Among these, analysis of demographics (age, gender, and race) has received tremendous attention owing to its wide real-world applications, with significant efforts devoted and great progress achieved. This survey first presents biometric demographic analysis from the standpoint of human perception, then provides a comprehensive overview of state-of-the-art advances in automated estimation from both academia and industry. Despite these advances, a number of challenging issues continue to inhibit its full potential. We second discuss these open problems, and finally provide an outlook into the future of this very active field of research by sharing some promising opportunities.

Recently, the electric vehicles (EV) are promoting rapidly in many big cities of China, but the charging problem has become a big obstacle for the development of EV. In this article, the demand of EV is determined on the basis of EV proportion and divided into 2 portions: (1) Centralized demand (2) Decentralized demand. The charging total mileage is determined based on the state of charge (SOC) and the orderly-free charging strategy is presented considering the demand of the users. Moreover, a new multi-objective teaching–learning based optimization (MOTLBO) algorithm is put forward to plan the quantity and location of the concentrated charging pile in the region considering the situation of the new city. Results showed that the performance of MOTLBO was promising and benefits can be achieved easily. As Compared to other methods, the method used in this article picks out the hierarchical planning idea firstly and addicts the man–machine interaction into the free-orderly charging strategy.

Medium-voltage distribution network faults are closely related to users' power consumption, so timely and effective early warning of distribution network faults is of great significance to ensure the safe and stable operation of the power grid. Aiming at the characteristics of distribution network, this paper presents a fault early warning method for medium voltage distribution network using "teaching and learning" to optimize non-linear state estimation (TLBO-NSET). Data related to faults are collected, and the data of the same time are regarded as a fault scenario state of the line. The historical data are sorted together to form a fault state set. For the future fault state, TLBO is used to optimize the middle weight of the nonlinear state estimation to predict the future fault level. The effectiveness of this method in distribution network early warning is validated by the analysis of 10 KV medium voltage line fault in a certain area in recent two years.

Based on the existing fault location of distribution network, a novel traveling wave head detection method based on MRSVD decomposition and symmetric difference energy operator is proposed in this paper. Multi-resolution singular value decomposition (MRSVD) is a method that combines matrix binary recursive structure principle and singular value decomposition method. In this paper, the original signal is separated from the high frequency detail components by MRSVD method, and then the energy of teager is calculated to obtain the precise traveling wave arrival time. Then, a multi-terminal traveling wave fault location scheme is proposed, and the fault location is obtained by layer by layer calculation. Finally, the accuracy of the algorithm is verified by MATLAB simulation. Experiments on setting different fault parameters show that the method has good accuracy and practicability.

Electricity consumption forecasting is an important issue in smart grid construction. Accurate electricity consumption forecasting has important guiding significance for power grid planning and economic sector management decision-making. This paper proposes a short-term electricity consumption forecasting method based on fuzzy clustering and backpropagation neural network algorithm using the electricity consumption data of residential users obtained by the measurement automation system every 15 minutes. The method first classifies the residential users according to the electricity usage behavior by clustering algorithm, and then takes the influence factors such as temperature, humidity, date type, wind speed and historical power consumption as training samples of backpropagation neural network, and performs short-term electricity consumption forecasting by type. The example analysis shows that the method is effective and feasible. Compared with the overall forecasting, the algorithm effectively reduces the forecasting error and improves the forecasting accuracy.

Due to the complexity of communication channel and nodes' communication range limit, the gateway nodes in power line communication network cannot directly communicate with all the child nodes. It is necessary to adopt the networking technology. Firstly, the gateway node establishes communication link to some child nodes, then these sub-nodes act as relays and continue to network, gradually expanding the communication range until all nodes join the network. In order to improve the reliability of power line communication, this paper proposes an improved clustering-cobweb algorithm based on the study of clustering-cobweb algorithm and the topology characteristics of low-voltage power line network. The improved algorithm adds an optimization step that each new node chooses the best cluster head node when joining the network. It can be proved through simulation that comparing with the original algorithm, the improved algorithm optimizes the network's structure and improves the reliability of the power line communication by reducing the average communication distance.

Taking a county mountain area in Guangdong Province as an example, this paper designs and implements a grid emergency repair management command system which integrates a website and a companion smartphone APP. The system adopts B/ S mode, asp.Net technology, and uses C# as the main development language, also combines Microsoft SQL Sever 2008 and other related software. Besides, it uses the method of mixed programming of MATLAB and C# to achieve the functions in distribution network, such as fault location, power quality analysis, fault early warning, etc. In addition, by taking advantage of the form of interaction between web page and APP, the system not only optimizes the process of work order but also improves the efficiency of emergency repair. The system has strong interactivity and maneuverability, and it also plays an important role in promoting the development of distribution automation in mountain areas.

In feature learning tasks, one of the most enormous challenges is to generate an efficient discriminative subspace. In this paper, we propose a novel subspace learning method, named recursive discriminative subspace learning with an l

Many face recognition systems have demonstrated promising results under well-controlled conditions with cooperative users. However, face recognition in real-world scenarios is still a challenging problem due to dramatic facial variations caused by different poses, lighting conditions, expressions, occlusion and so on. In this chapter, we summarize recent advances in unconstrained face recognition. We begin by introducing existing unconstrained face databases or benchmarks. We then provide an overview of recent techniques specifically developed for this task, including advanced face representations, metric learning approaches, background information investigation and pose-invariant approaches. Finally, we highlight some open issues to be addressed.

© 2014 IEEE.Facial imaging has been largely addressed for automatic personal identification, in a variety of different environments. However, automatic face recognition becomes very challenging whenever the acquisition conditions are unconstrained. In this paper, a picture-specific cohort normalization approach, based on polynomial regression, is proposed to enhance the robustness of face matching under challenging conditions. A careful analysis is presented to better understand the actual discriminative power of a given cohort set. In particular, it is shown that the cohort polynomial regression alone conveys some discriminative information on the matching face pair, which is just marginally worse than the raw matching score. The influence of the cohort set size in the matching accuracy is also investigated. Further, tests performed on the Face Recognition Grand Challenge ver 2 database and the labeled faces in the wild database allowed to determine the relation between the quality of the cohort samples and cohort normalization performance. Experimental results obtained from the LFW data set demonstrate the effectiveness of the proposed approach to improve the recognition accuracy in unconstrained face acquisition scenarios.

This paper presents a novel method which takes advantage of the K-means clusters and effective degree of clustering to recognize the abnormal data of electricity consumption by PLC communication. The novel clustering method was used to supervise the abnormal or missing data by the badly quality of PLC communication effectively. Firstly, we extract good historical data as the basic cluster according to the selected characteristic numbers. Then, the current moment of electricity consumption data take clusters analysis with the basic cluster to identify whether the data is abnormal or not. Finally, the paper conducted simulation to verify the method and the result of simulation showed that the method can verify the abnormal data effectively. Therefore, the method is beneficial for the supervision for data.

OFDM (Orthogonal Frequency Division Multiplexing) system is one of the most widely used power line communication (PLC) systems, and the noise of power line channel is one of the main factors that affects the reliability of power line communication. The development of power line communication technology is of great practical significance for the remote monitoring and remote meter reading of power system, and it plays an irreplaceable role in the development of smart grid. In this paper, the OFDM system model of data flow were set up on the basic principle of OFDM communication system, and the circuits are also set up to measure the noise on the power line. According to the measured data, the power line colored background noise model, narrowband noise model and burst noise model are set up respectively. For each noise model the calculations and simulations have been carried individually in the paper. Finally, the simulations by changing the encoding method and SNR (Signal to Noise Ratio) of the OFDM system have been carried. Eventually the effect of encoding and SNR of the OFDM system for the BER (Bite Error Rate) and the results of simulations by curve graphs are presented. In the end of the paper, some advice to improve the quality of PLC in different noise environment based on the experimental results are giving. The paper may be useful to the studies of PLC and the optimization of power system.

Nowadays, surveillance cameras are widely installed in public places for security and law enforcement, but the video quality may be low because of the limited transmission bandwidth and storage capacity. In this study, the authors proposed a gait recognition method for extremely low-quality videos, which have a frame-rate at one frame per second (1 fps) and resolution of 32 × 22 pixels. Different from popular temporal reconstruction-based methods, the proposed method uses the average gait image (AGI) over the whole sequence as the appearance-based feature description. Based on the AGI description, the authors employed a large number of weak classifiers to reduce the generalisation errors. The performance can be further improved by incorporating the model-based information into the classifier ensemble. The authors found that the performance improvement is directly proportional to the average disagreement level of weak classifiers (i.e. diversity), which can be increased by using the model-based information. The authors evaluated the proposed method on both indoor and outdoor databases (i.e. the low-quality versions of OU-ISIR-D and USF databases), and the results suggest that our method is more general and effective than other state-of-the-art algorithms.

As the growth of the operating life and the influence of the environment, the hydrophobic of transformer composite insulator bushing will decline and pollution flashover accident happened easily, which can affect the safe operation of lines. This paper using machine vision to monitor the transformer composite insulator bushing surface, first, recognition and preprocessing the images of hydrophobic of composite insulator bushing, then extracting the characteristic value, finally, using probability neural network (PNN) to evaluate the hydrophobic of the transformer composite insulator bushing, after a large number of samples training, PNN model can accurately distinguish the hydrophobic level of bushing composite insulator bushing, which has great significance for monitoring composite insulator reliably.

As all know, the increasing of filth on the Composite insulator bushing will lead to drop the hydrophobicity and insulating properties of the bushing. According to the relationship between the degree of the filth on the composite insulator bushing and the hydrophobicity, we can use hydrophobicity detection to react the extent of the filth on the bushing. This paper using digital image processing technology to process the collected droplet photo of the composite insulator bushing, and then extract the four characteristic values from the processed photo. This study inputs 420 groups of samples to the Multilayer Perceptron Neural Network (MLP) model for training, and the MLP is improved, the detection effect compared MLP with the improved MLP by 210 test samples, it shows that the improved MLP accuracy is higher, which can be better used in the detection of composite insulator bushing hydrophobic and can reflect the degree of the filth on the composite insulator bushing better. The study of the paper has important significance on the clean-up of composite insulator bushing.

According to the charging demand of electric vehicle and the power that charging station can provide, this paper pick up a Human–Computer Interacted Strategy which divides the charging process into 4 phases and proposes the modes of charging for guiding the users to take full advantage of electric vehicle charging pile. Based on the queuing theory, reverse order queuing and mixed order queuing, this paper determines the sequence of electric vehicle charging plan in off-peak phases. Comparing with disordered charging strategy and existing charge strategy, the strategy in this paper can make the charging station satisfying more demand and getting more benefits. Moreover, the strategy also has good performance in valley filling for the power system.

According to the randomness and fuzziness of electric vehicle, this paper determines the conception of membership by the fuzzy statistical, introduces the theory of membership cloud model and builds the electric vehicle charging load model. Respectively, this paper describes the electric car charging load in different stage, constructs the whole model of electric vehicle charging load demand and determines the range of the change of the electric vehicle charging load by cloud model. Combined with a large number of actual statistical data, the result was compared with Monte Carlo method and the accuracy of the proposed model is verified. The proposed model is suitable for all kinds of charging load prediction for specific areas and facilitates the predicted results applying in related electric cars infrastructure planning.

Face pair matching is the task of determiningwhether two face images represent the same person. Due tothe limited expressive information embedded in the two faceimages as well as various sources of facial variations, it becomesa quite difficult problem. Towards the issue of few availableimages provided to represent each face, we propose to exploit anextra cohort set (identities in the cohort set are different fromthose being compared) by a series of cohort list comparisons.Useful cohort coefficients are then extracted from both sortedcohort identities and sorted cohort images for complementaryinformation. To augment its robustness to complicated facialvariations, we further employ multiple face modalities owingto their complementary value to each other for the face pairmatching task. The final decision is made by fusing the extractedcohort coefficients with the direct matching score for all theavailable face modalities. To investigate the capacity of eachindividual modality on matching faces, the cohort behavior andthe performance achieved by using our complementary cohortstrategy, we conduct a set of experiments on two recently collectedmultimodal face databases. It is shown that using differentmodalities leads to different face pair matching performance. Foreach modality, employing our cohort scheme significantly reducesthe equal error rate. By applying the proposed multimodalcomplementary cohort strategy, we achieve the best performanceon our face pair matching task.

The normal communication between smart meter and concentrator is a key factor influencing the normal function of users&rsquo, power consumption systems. To solve the communication failure of the smart meter caused by the signal conflict as well as the collected consecutive information abnormality from the same smart meter, according to the chain optimization index, the networking method of static and dynamic combination proposed in this paper is first used to picked out the optimal relay for a smart meter belonging to multiple relay communication ranges. Meanwhile, the communication with other secondary relays is closed to avoid signal conflict. Then the paper forms different combinations of collected data and these combinations are trained in the extreme learning machine (ELM) to find the characteristics value of power consumption information. Finally, in MATLAB simulation, if ELM detects the abnormal information, new communication path could be promptly found through dynamic adjustment of chain optimization weighted coefficient and the weighted coefficient of the number of the relayed smart meters. It solves the problem of consecutive information abnormality from the same smart meter and raises the reliability of smart meter&rsquo, s communication, having a significantly meaning to guarantee the normal function of users&rsquo, power consumption system.