Search

The task of text-to-image synthesis is a new challenge in the field of image synthesis. In the earlier research, the task of text-to-image synthesis is mainly to achieve the alignment of words and images by the way of retrieval based on the sentences or keywords. With the development of deep learning, especially the application of deep generative models in image synthesis, image synthesis achieves promising progress. The Generative adversarial networks (GANs) are one of the most significant generative models, and GANs have been successfully applied in computer vision, natural language processing and so on. In this paper, we review and summarize the recent research in GANs-based text-to-image synthesis, and provide a summary of the development of classic and advanced models. The input of the GANs-based text-to-image synthesis is not only the general text description as earlier studies, also includes scene layout and dialog text. The typical structure of each categories is elaborated. The general text-based image synthesis is the most commonly in the text-to-image synthesis, and it is subdivided into three groups based on the improvements of text information utilization, network structure and output control conditions. Through the survey, the detailed and logical overview of the evolution of GANs-based text-to-image synthesis is presented. Finally, the challenged problems and the future development of text-to-image synthesis are discussed.

The dual stream-based human action recognition model offers the advantage of high recognition accuracy, but the algorithm is less robust in case of lighting changes. The human skeleton has a strong ability to express human behavior and actions; however, the scene information is ignored. Drawing on the idea of the dual-stream model, this paper proposes a human skeleton and scene image-based dual-stream model for human action recognition. The motion features are extracted through the spatio-temporal graph convolution of the human skeleton, and a scene recognition model is proposed based on the sparse frame sampling of video and video-level consensus strategy to process the scene video and gather the visual scene information. The proposed model exploits the advantages of skeleton information in motion expression and the superiority of the image in scene presentation. The scene information and spatio-temporal graph convolution-based human skeleton limbs are fused complementarily to achieve human action recognition. Compared to the conventional optical flow-based dual-stream action recognition method, this model is verified by experimenting under unstable light conditions, and the performance of human action recognition is robust and promising.

There have been several attempts to integrate a spatial visual attention mechanism into an image caption model and introduce semantic concepts as the guidance of image caption generation. High-level semantic information consists of the abstractedness and generality indication of an image, which is beneficial to improve the model performance. However, the high-level information is always static representation without considering the salient elements. Therefore, a semantic attention mechanism is used for the high-level information instead of conventional of static representation in this article. The salient high-level semantic information can be considered as redundant semantic information for image caption generation. Additionally, the generation of visual words and non-visual words can be separated, and an adaptive attention mechanism is employed to realize the guidance information of image caption generation switching between new fusion information (fusion of image feature and high-level semantics) and a language model. Therefore, visual words can be generated according to the image features and high-level semantic information, and non-visual words can be predicted by the language model. The semantics attention, adaptive attention, and previous generated words are fused to construct a special attention module for the input and output of long short-term memory. An image caption can be generated as a concise sentence on the basis of accurately grasping the rich content of the image. The experimental results show that the performance of the proposed model is promising for the evaluation metrics, and the captions can achieve logical and rich descriptions.

AI technology has developed so fast, and it has been applied to the commercial area. In order to predict the customer preference and adjust the placement of product or advertisement, etc., the inte...

At present, China's emerging technologies enjoy vigorous development, and emerging industrial chain with artificial intelligence and big data as the mainstay has dominated China's new economic form. With it, the question of how to cultivate innovative and compound talents in emerging disciplines has become one core difficulty in the current higher education. The highly interdisciplinary integrated STEAM education is suitable for the initial training of new talents. This paper studies the status quo of STEAM education and artificial intelligence education in Chinese universities. Based on the practice of STEAM talent training in Anshan Normal University, it combines artificial intelligence-related education practice, tracks and analyzes various indicators of students in STEAM course to explore the feasibility plan for artificial intelligence education in Chinese universities.

Energy conservation, environmental protection, and intelligence are topics of interest in intelligent buildings. However, the energy requirement of various electrical equipment in intelligent buildings increases energy consumption. This study presents a neural network-based prediction and control system for the regulation of building environmental parameters. Neural network-based soft sensing technology can detect building environmental parameters through few sensors. The proposed system control algorithm can realize the adaptive adjustment of environmental parameters by using a neural network proportional–integral–derivative controller. Zigbee wireless communication is adopted as the information transmission medium to realize the environmental parameter measurement and network control. The soft sensing technique combined with Zigbee communication technology can effectively reduce energy consumption. The central control system analyzes the data coming from the network and regulates the environmental parameter through lifting temperature, ventilation, and switching curtains by using the neural network proportional–integral–derivative algorithm. The regulation of environmental parameters reduces unnecessary energy consumption. Finally, the effectiveness of the system is verified through simulations. Practical applications: This work reports an energy saving scheme. The building communication system constructed by ZigBee can reduce energy consumption and can be easily expanded. The soft sensing technique based on artificial neural network can predict temperatures by using few sensors. The neural network proportional–integral–derivative control algorithm has good performance in the regulation of environmental parameters for a time-varying system. Building energy consumption can be reduced by conducting these measures.

In the practical scene, object detection faces a very complicated situation. The occlusion problem always occurs in actual scene, which may affect the accuracy of object detection, especially for the occluded objects. For the deep models, a larger dataset with sufficient occlusion samples will improve the performance of the object detection models. However, the sample with occlusion problem is too hard to obtain. Therefore, a global average pooling(GAP) based adversarial Faster-RCNN is proposed to generate the hard samples and enhance the performance of object detection algorithm. Sufficient hard samples can be generated with the help of this model. Therefore, the object detection model can be trained adequately for the occluded objects. The hard sample generation is carried out in the space of image feature instead of image generation directly. The class-dependent part is obtained by the GAP network, and it is obscured to generate the feature map of hard sample for model reinforcement training. Therefore, the better object detection model can be trained using a conventional dataset. The Faster-RCNN is adopted as the baseline. The Faster-RCNN and GAP have a joint training to improve the performance of the proposed model. The simulation results exhibit the validation of the proposed algorithm.

Development of curative treatments for glioblastoma (GBM) has been stagnant in recent decades largely because of significant financial risks. A portfolio-based strategy for the parallel discovery of breakthrough therapies can effectively reduce the financial risks of potentially transformative clinical trials for GBM. Using estimates from domain experts at the National Brain Tumor Society (NBTS), we analyze the performance of a portfolio of 20 assets being developed for GBM, diversified across different development phases and therapeutic mechanisms. We find that the portfolio generates a 14.9% expected annualized rate of return. By incorporating the adaptive trial platform GBM AGILE in our simulations, we show that at least one drug candidate in the portfolio will receive US Food and Drug Administration (FDA) approval with a probability of 79.0% in the next decade.

Recently, the information analysis technology of underwater has developed rapidly, which is beneficial to underwater resource exploration, underwater aquaculture, etc. Dangerous and laborious manual work is replaced by deep learning-based computer vision technology, which has gradually become the mainstream. The binocular cameras based visual analysis method can not only collect seabed images but also construct the 3D scene information. The parallax of the binocular image was used to calculate the depth information of the underwater object. A binocular camera based refined analysis method for underwater creature body length estimation was constructed. A fully convolutional network (FCN) was used to segment the corresponding underwater object in the image to obtain the object position. A fish&rsquo, s body direction estimation algorithm is proposed according to the segmentation image. The semi-global block matching (SGBM) algorithm was used to calculate the depth of the object region and estimate the object body length according to the left and right views of the object. The algorithm has certain advantages in time and accuracy for interest object analysis by the combination of FCN and SGBM. Experiment results show that this method effectively reduces unnecessary information, improves efficiency and accuracy compared to the original SGBM algorithm.

In this work, we propose a generic visual analytics framework to support tactic analysis based on data collected from racquet sports (such as tennis and badminton). The proposed approach models each rally in a game as a sequence of hits (i.e., events) until one athlete scores a point. Each hit can be described with a set of attributes, such as the positions of the ball and the techniques used to hit the ball (such as drive and volley in tennis). Thus, the mentioned sequence of hits can be viewed as a multivariate event sequence. By detecting and analyzing the multivariate subsequences that frequently occur in the rallies (namely, tactical patterns), athletes can gain insights into the playing styles adopted by their opponents, and therefore help them identify systematic weaknesses of the opponents and develop counter strategies in matches. To support such analysis effectively, we propose a steerable multivariate sequential pattern mining algorithm with adjustable weights over event attributes, such that the domain expert can obtain frequent tactical patterns according to the attributes specified by himself. We also propose a re-configurable glyph design to help users simultaneously analyze multiple attributes of the hits. The framework further supports comparative analysis of the tactical patterns, e.g., for different athletes or the same athlete playing under different conditions. By applying the framework on two datasets collected in tennis and badminton matches, we demonstrate that the system is generic and effective for tactic analysis in sports and can help identify signature techniques used by individual athletes. Finally, we discuss the strengths and limitations of the proposed approach based on the feedback from the domain experts.

In the midst of epidemics such as COVID-19, therapeutic candidates are unlikely to be able to complete the usual multiyear clinical trial and regulatory approval process within the course of an outbreak. We apply a Bayesian adaptive patient-centered model—which minimizes the expected harm of false positives and false negatives—to optimize the clinical trial development path during such outbreaks. When the epidemic is more infectious and fatal, the Bayesian-optimal sample size in the clinical trial is lower and the optimal statistical significance level is higher. For COVID-19 (assuming a static R_{0} = 2 and initial infection percentage of 0.1%), the optimal significance level is 7.1% for a clinical trial of a nonvaccine anti-infective therapeutic and 13.6% for that of a vaccine. For a dynamic R_{0} decreasing from 3 to 1.5, the corresponding values are 14.4% and 26.4%, respectively. Our results illustrate the importance of adapting the clinical trial design and the regulatory approval process to the specific parameters and stage of the epidemic.

In the midst of epidemics such as COVID-19, therapeutic candidates are unlikely to be able to complete the usual multi-year clinical trial and regulatory approval process within the course of an outbreak. We apply a Bayesian adaptive patient-centered model—which minimizes the expected harm of false positives and false negatives—to optimize the clinical trial development path during such outbreaks. When the epidemic is more infectious and fatal, the Bayesian-optimal sample size in the clinical trial is lower and the optimal statistical significance level is higher. For COVID-19 (assuming a static R0 = 2 and initial infection percentage of 0.1%), the optimal significance level is 7.1% for a clinical trial of a non-vaccine anti-infective therapeutic clinical trial and 13.6% for that of a vaccine. For a dynamic R0 ranging from 2 to 4, the corresponding values are 14.4% and 26.4%, respectively. Our results illustrate the importance of adapting the clinical trial design and the regulatory approval process to the specific parameters and stage of the epidemic.

Aiming at the problems of rolling bearings fault diagnosis, this paper proposes a hybrid intelligent fault diagnostic model which combines stacked sparse auto-encoding network (SSAE) and KNN classifier. As a branch of deep learning, SSAE has excellent ability of feature learning, which is able to reduce the dimension of frequency domain signal and extract deep features. With the help of sparse constraint, a penalty term is added to strengthen its generalization ability. Combined with the advantages of K-Nearest Neighbor algorithm (KNN) in dealing with multi-classification problems, the accuracy of the proposed model can be improved. In addition, for the purpose of validating the feasibility and superiority of the method, extensive contrast experiments are carried out. Experimental results show that compared to traditional methods, making use of deep neural network for self-extraction frequency-domain features not only avoids excessive dependence on professional knowledge and project experience, but also improves the accuracy of fault classification.

Image captioning is a hot topic of image understanding, and it is composed of two natural parts ("look" and "language expression") which correspond to the two most important fields of artificial intelligence ("machine vision" and "natural language processing"). With the development of deep neural networks and better labeling database, the image captioning techniques have developed quickly. In this survey, the image captioning approaches and improvements based on deep neural network are introduced, including the characteristics of the specific techniques. The early image captioning approach based on deep neural network is the retrieval-based method. The retrieval method makes use of a searching technique to find an appropriate image description. The template-based method separates the image captioning process into object detection and sentence generation. Recently, end-to-end learning-based image captioning method has been verified effective at image captioning. The end-to-end learning techniques can generate more flexible and fluent sentence. In this survey, the image captioning methods are reviewed in detail. Furthermore, some remaining challenges are discussed.

The deep learning techniques have received great achievements in computer vision, natural language processing, etc. The success of deep neural networks depends on the sufficient training of parameters. The traditional way of neural network training is a gradient-based algorithm, which suffers the disadvantage of gradient disappearing, especially for the deeper neural network. Recently, a heuristic algorithm has been proposed for deeper neural network optimization. In this paper, a random mask and elitism univariate continuous estimation of distribution algorithm based on the Gaussian model is proposed to pre-train staked auto-encoder, and then a Stochastic Gradient Descent (SGD) based fine-tuning process is carried out for local searching. In the improved estimation of the distribution algorithm, two individual update strategies are defined; one group of individuals is generated according to the constructed probabilistic model, and another is updated according to the statistics of advanced individuals that aim to reduce the probability of combination explosion and time consumption according to the mask information. In the simulations, different architectures, different mask ratios and different promising individual ratios are adopted to testify the effectiveness of the improved algorithm. According to simulation results, the estimation of thr distribution algorithm has a steady optimization ability for the shallow and stacked auto-encoder by one-step pre-training combining SGD based fine-tuning for the MNIST dataset. The proposed model will achieve a state-of-the-art performance on Fashion-MNIST.

The automation level of autonomous marine vehicle is limited which is always semi-autonomy and reliant on operator interactions. In order to improve it, an autonomous collision avoidance method is proposed based on the visual technique as human’s visual system. A deep convolutional neural network (Alexnet), with strong visual processing capability, is adopted for encounter pattern recognition. European Ship Simulator is used to generate some encounter scenes and record the corresponding maneuver operation conforming to the COLREGs (International Regulations for Preventing Collisions at Sea) rules as samples. After the training phase, of Alexnet, it can successfully predict the collision avoidance operation according to the input scene image like crewman; moreover, this can provide operation guidance for the automatic navigation, guidance and control system. Some different encounter situations are simulated, and used to testify the validity of the proposed approach.

Compared with the achievements of convolutional neural networks in image classification, human action recognition for video is not ideal in terms of accuracy and practicability. A major method in action recognition is based on the human skeleton, which is an important information for characterizing human motion in video. In this paper, the human skeleton in video is extracted by OpenPose, and the spatial and temporal graph of skeleton is constructed. The spatial and temporal graph convolution network (ST-GCN) is used to extract the spatial and temporal features of the human skeleton on consecutive video frames, and the features is used for video classification. In order to verify the action recognition performance based on the ST-GCN, a 50.53% top-1 and 81.58% top-5 accuracy is obtained on the UCF-101 dataset. A specific UCF-31 dataset is constructed manually and a 68.73% top-1 and 94.43% top-5 accuracy is obtained, verifying that the identification accuracy of ST-GCN model would also be improved when the accuracy of skeleton acquisition was improved.

Scene perception of mobile robot is that the robot realizes the perception and understanding of the surrounding environment through a series of sensors configured by itself, and the perception technology based on vision has always been a hot and difficult point in research. Therefore, in terms of visual environment perception algorithms, there have been numerous valuable research achievements in recent years. In particular, the object detection and segmentation algorithm based on convolutional neural network has shown good performance in simple scene, but there are still some limitations when these algorithms are directly applied to actual scenes. In this paper, we study the practical application of vision-based environmental perception of mobile robots in complex scenes, presents a unified algorithm architecture of object detection and road segmentation, and build a vision-based mobile robot's environment perception system. Firstly, the image acquisition of the surrounding environment is completed by the computer camera mounted on the robot, and then the obstacle detection and the segmentation of the drivable area are achieved by using the target detection and segmentation algorithm. In order to meet the real-time requirements, the detection and segmentation algorithms share the same feature extraction network, and are jointly trained as one framework. Finally, according to the detection and segmentation results, the robot can automatically avoid obstacles and move in the drivable area.

This proposal expresses a standpoint concerning a discussion for the phenomenon of a poor sense of copyright in Chinese youth netizens and the suggestion. The phenomenon of a poor sense of copyright in Chinese youth netizens has disadvantages. By analyzing it, a clear status of Chinese youth netizens' problems on the poor sense of copyright will be shown. At the same time, when we consider a discussion for the phenomenon of a poor sense of copyright in Chinese youth netizens and the suggestion, we must address several main reasons that contribute to problem. These include cultural tradition restriction, incomprehensive market system, inappropriate price incentive, imperfective legal institutions and indifference education system. It is valuable for the originators feel free to create better works. Meanwhile, those factors are essential for Chinese youth netizens to consider the way to be cultivated the right spending habits. Only by analyzing all these factors together and then using these results to apply to intensity notional cultural competitiveness and form an innovative market. This paper is organized as follows: The first part will outline the poor sense of copyright in Chinese youth netizens' case studies and analyze. The second part is an introduction of the problems of the phenomenon of a poor sense of copyright in Chinese youth netizens. The reasons causing the problems are analyzed in part three. Some proposals to overcome copyright problem will be advocated in last part. This study found that enhancing copyright awareness is a very complex task for Chinese youth netizens. Cultural tradition restriction, incomprehensive market system, inappropriate price incentive, imperfective legal institutions and indifference education system are all factors which bring difficulties to promote copyright awareness in Chinese youth netizens. Combined those factors with four problems of the phenomenon in China, I offer appropriate suggestions to overcome problems and help youth netizens acquire the knowledge of copyright.

Surveillance video analysis has applied to kinds of areas. Mixed GMM is a traditional target separation method. In order to overcome the object detection performance under dynamic scene, an improved algorithm is adopted to improve the mixed Gaussian model algorithm. A different initialization and model parameter updating strategy are adopted to detect the object under dynamic scene. The experiments validate the effectiveness of the algorithm at the lakeside and geyser.

Collaborative control of a dual-arm robot refers to collision avoidance and working together to accomplish a task. To prevent the collision of two arms, the control strategy of a robot arm needs to avoid competition and to cooperate with the other one during motion planning. In this paper, a dual-arm deep deterministic policy gradient (DADDPG) algorithm is proposed based on deep reinforcement learning of multi-agent cooperation. Firstly, the construction method of a replay buffer in a hindsight experience replay algorithm is introduced. The modeling and training method of the multi-agent deep deterministic policy gradient algorithm is explained. Secondly, a control strategy is assigned to each robotic arm. The arms share their observations and actions. The dual-arm robot is trained based on a mechanism of “rewarding cooperation and punishing competition”. Finally, the effectiveness of the algorithm is verified in the Reach, Push, and Pick up simulation environment built in this study. The experiment results show that the robot trained by the DADDPG algorithm can achieve cooperative tasks. The algorithm can make the robots explore the action space autonomously and reduce the level of competition with each other. The collaborative robots have better adaptability to coordination tasks.

In view of difficulty in application of optical flow based human action recognition due to large amount of calculation, a human action recognition algorithm I3D-shufflenet model is proposed combining the advantages of I3D neural network and lightweight model shufflenet. The 5 &times, 5 convolution kernel of I3D is replaced by a double 3 &times, 3 convolution kernels, which reduces the amount of calculations. The shuffle layer is adopted to achieve feature exchange. The recognition and classification of human action is performed based on trained I3D-shufflenet model. The experimental results show that the shuffle layer improves the composition of features in each channel which can promote the utilization of useful information. The Histogram of Oriented Gradients (HOG) spatial-temporal features of the object are extracted for training, which can significantly improve the ability of human action expression and reduce the calculation of feature extraction. The I3D-shufflenet is testified on the UCF101 dataset, and compared with other models. The final result shows that the I3D-shufflenet has higher accuracy than the original I3D with an accuracy of 96.4%.

Adaptive Resonance Theory (ART) model is a special neural network based on unsupervised learning which simulates the cognitive process of human. However, ART1 can be only used for binary input, and ART2 can be used for binary and analog vectors which have complex structures and complicated calculations. In order to improve the real-time performance of the network, a minimal structural ART is proposed which combines the merits of the two models by subsuming the bottom-up and top-down weight. The vector similarity test is used instead of vigilance test. Therefore, this algorithm has a simple structure like ART1 and good performance as ART2 which can be used for both binary and analog vector classification, and it has a high efficiency. Finally, a gas turbine fault diagnosis experiment exhibits the validity of the new network.

Video1 surveillance has been applied to various industries and business, but the video surveillance is mainly to provide video capture and display services at current, and does not care about video content. In some applications, users want to be able to monitor the video content and implement automatic prompting and alarming functions. Therefore, this paper uses the hybrid Gaussian model (GMM) to achieve the separation of the foreground and the background of the video and the experiments validate the effectiveness of the algorithm.

Fully convolutional neural network is a special deep neural networks based on convolutional neural networks and are often used for semantic segmentation. This paper proposes an improved fully convolutional neural network which fuses the feature maps of deeper layers and shallower layers to improve the performance of image segmentation. In the process of feature fusion, adaptive parameters are introduced to enable different layers to participate in feature fusion as different proportion. The deep layers of neural network mainly extract the abstract information of the object, and the shallow layers of neural network extracts the refined features of objects, such as edge information and precise shape. Adaptive parameters can speed up the training speed and improve the prediction accuracy. In the early stages of training, the feature maps of shallow layers have a larger fusion coefficient, which allows the neural network to learn the feature of object's location and shape quickly. As the training progresses, gradually weakening the fusion coefficient of shallow layers and increasing the fusion coefficient of deep layers which can enhance the network's ability of predicting the details of the objects. This paper uses Scene Parsing Challenge 2016 dataset presented by MIT for training. Experiments show that the method proposed in this paper speeds up the training and improves the pixel prediction accuracy.

Depth estimation plays an essential part in understanding the three-dimensional (3D) geometric relations of a scene. Compared with other methods such as binocular vision, estimating depth from monocular image is much more challenging. In this paper, we propose a conditional generative adversarial net (cGAN) to tackle the problem of monocular image depth estimation. For enhancing the learning of our net in the training phrase, cycle consistency is applied to our network to form a closed loop. We use the network to model the mapping between the RGB images domain and the depth images domain. After training the network adequately, the model can output depth image according to the input RGB image. Experiments on NYU Depth v2 dataset demonstrate the proposed method outperforms state-of-art depth estimation approaches.

In recent years, the robotics industry has developed rapidly. Wheeled service robots have been used in many fields. This paper constructs a wheeled mobile robot based on two-wheeled drive, and designs the robot motion controller to realize the motion control of the robot. The control system includes a pair of DC motors, a driver board, a STM32 controller, and two Hall sensors. The Hall sensors are used for measuring the motor speed to realize a speed closed loop control. The PID algorithm is adopted to control the speed of two DC motors to achieve precise control of the speed of the mobile platform. The experiments exhibit that the wheeled mobile robot can move flexibly, such as straight-going or turning around.

The electric motor is the elementary device of modern industry system, and its timely fault diagnosis leads to reduce the maintenance costs and downtime, and improve system reliability. This paper deals with the problem of fault diagnosis of electric motor based on power signal, and a data-driven fault diagnosis method based on genetic algorithm (GA) optimized support vector machine (SVM) is presented. The feature presentation, feature selection and feature extraction are applied as data preprocessing methods to reduce data dimensions, that is, we implement feature representation by the time domain analysis method and the range analysis method, and the fisher discriminant analysis is used for feature selection, and the locally linear embedding (LLE) is used for feature extraction. Then the GA is used to optimize the SVM classifier for fault classification after data preprocessing. Our method can obtain good fault classification effect, and the experimental results show that the classification accuracy of the proposed method is better than that of probabilistic neural network, and the feasibility and effectiveness of this proposed method in fault diagnosis of electric motor are proved.

Distant metastasis is the main cause of breast cancer-related death; however, effective therapeutic strategies targeting metastasis are still scarce. This is largely attributable to the spatiotemporal intratumor heterogeneity during metastasis. Here we show that protein deacetylase SIRT7 is significantly downregulated in breast cancer lung metastases in human and mice, and predicts metastasis-free survival. SIRT7 deficiency promotes breast cancer cell metastasis, while temporal expression of Sirt7 inhibits metastasis in polyomavirus middle T antigen breast cancer model. Mechanistically, SIRT7 deacetylates and promotes SMAD4 degradation mediated by β-TrCP1, and SIRT7 deficiency activates transforming growth factor-β signaling and enhances epithelial-to-mesenchymal transition. Significantly, resveratrol activates SIRT7 deacetylase activity, inhibits breast cancer lung metastases, and increases survival. Our data highlight SIRT7 as a modulator of transforming growth factor-β signaling and suppressor of breast cancer metastasis, meanwhile providing an effective anti-metastatic therapeutic strategy., Metastatic disease is the major reason for breast cancer-related deaths; therefore, a better understanding of this process and its players is needed. Here the authors report the role of SIRT7 in inhibiting SMAD4-mediated breast cancer metastasis providing a possible therapeutic avenue.

Unsupervised learning is a good neural network training way. However, the unsupervised learning algorithm is rare. The generative model is an interesting algorithm which can generate the similar data as the sample data by building a probabilistic model of the input data, and it can be used for unsupervised learning. Variational autoencoder is a typical generative model which is different from common autoencoder that a probabilistic parameter layer follows the hidden layer. Some new data can be reconstructed according to probabilistic model parameters. The probabilistic model parameter is the latent variable. In this paper, we want to do some research to test the data reconstruct effect of the variational autoencoder by different latent variables. According to the simulation, the more latent variables the more style of the sample is.

Autoencoder is an excellent unsupervised learning algorithm. However, it can not generate kinds of sample data in the decoding process. Variational autoencoder is a typical generative adversarial net which can generate various data to augment the sample data. In this paper, we want to do some research about the information learning in hidden layer. In the simulation, we compare the hidden layer learning of hidden layer in conventional autoencoder and variational autoencoder.

We use danger immune algorithm to optimize ship collision avoidance strategy.Collision avoidance operations are encoded as the individuals of optimization algorithm.The collision avoidance theory and ship motion mathematical model are used.The system will output a safe and economical collision avoidance strategy. Marine collision accidents cause a great loss of lives and property. As a possible solution, the danger immune algorithm is used to achieve ship collision avoidance strategy optimization, which is a multi-objective problem concerning safety and economy. Collision avoidance operations are encoded as the individuals of optimization algorithm. In the system, ship domain and ship arena, among others, are used for collision risk evaluation to assess the fitness of individuals. Through the optimization, the navigator will obtain the optimal collision avoidance strategy to achieve safe and efficient collision avoidance. The simulations indicate that the optimization algorithm is valid.

The traditional immune algorithm (IA) is based on a self-nonself biological immunity mechanism. Recently, a novel immune theory called the danger model theory has provided more suitable biological information for data handling compared with the self-nonself mechanism. According to the danger model theory and based on past experiences of the genetic and artificial IA, we present the Danger Model Immune Algorithm (DMIA) that differs from the traditional IA in terms of the self-nonself biological immunity mechanism. We define a danger area and a danger signal in DMIA. We use the selection, mutation, and specific danger operators to update the population. The algorithm can achieve complex problem optimization. Simulation studies demonstrate that DMIA exhibits a higher efficiency than traditional genetic algorithms and other algorithms when considering a number of complicated functions.

The suitability evaluation has an important influence on suitable-matching region selection of geomagnetism matching navigation. Geomagnetic anomalies have a fractal characteristic. Fractal dimension can describe the self-similar characteristics and subtle changes of irregular, broken, uneven and infinite detail form of geomagnetic anomalies. A comprehensive evaluation function based on fractal characteristics was established, and the ranks of region suitability were gotten finally.

Condenser is an essential equipment of steam turbine. It is valuable to diagnose the fault when monitor the condenser system. To overcome shortcomings of BP neural network in fault diagnosis, ART1 (adaptive resonance architecture 1) is used to diagnose the condenser fault. The ART1 Neural network studies the typical fault features data, and realizes fault diagnosis. The results indicate the improved ART1 neural network can diagnose the condenser system fault correctly, and overcome the shortcoming of the original training algorithm. Although this is an application based on condenser, the algorithm can be used to further industrial system.

Stacked autoencoder is a typical deep neural network. The hidden layers will compress the input data with a better representation than the raw data. Stacked autoencoder has several hidden layers. However, the number of hidden layers is always experiential. In this paper, different hidden layers number autoencoders are discussed. Different depths of stacked autoencoder have different learning capability. The deeper stacked autoencoders have better learning capability which needs more training iterations and time.

Deep neural network (DNN) has many advantages. Autonomous driving has become a popular topic now. In this paper, an improved stack autoencoder based on the deep learning techniques is proposed to learn the driving characteristics of an autonomous car. These techniques realize the input data adjustment and solving diffusion gradient problem. A Raspberry Pi and a camera module are mounted on the top of the car. The camera module provides the images needed for training the DNN. There are two stages in the training. In the pre-training process, an improved autoencoder is trained by the unsupervised learning mechanism, and the characterization of the track is extracted. In the fine-tuning stage, the whole network is trained according to the labeled data, and then this model learns the driving characteristics better according to the samples. In the experimental stage, the car will predict the action of the car by the trained model in the autonomous mode. The experiment exhibits the effectiveness of the proposed model. Compared with the traditional neural network, the improved stack autoencoder has a better generalization ability and faster convergence speed.

Most of the existing typical ship domains have been comprehensively reviewed and classified. Most of these ship domains are described in a geometrical manner that is difficult to apply to practices and simulations in marine traffic engineering. According to different types of geometrical ship domains, we have proposed mathematical models, based on which a unified analytical framework has been established. It is feasible and practical for the analytical models to be applied to the assessment of navigational safety, collision avoidance and trajectory planning, etc. Finally, some computer simulations and comparative studies of the proposed domain model have been presented and the simulation results show that the uniform analytical framework for ship domains is effective and identical to the original geometrical ones. It should be noted that the analytical domain models could be directly applied in any collision risk, collision avoidance or VTS system while the geometrical ones would be more illustrative but less practical or analytical.

The viscoelastic multi-dimensional earthquake isolation and mitigation device is a new kind of passive control device, which has both earthquake isolation and earthquake mitigation abilities. In order to quantify the horizontal earthquake isolation and mitigation effect of this device on structures, shaking table tests on structures with and without the devices and the corresponding analysis on earthquake isolation and mitigation properties are carried out. It can be shown from the experimental and analytical results that the device has both earthquake isolation and earthquake mitigation effects on structures in horizontal direction, and its horizontal earthquake mitigation coefficients can be referred to those of the rubber bearing earthquake isolation structure.

Convolutional Neural Network (CNN) is a kind of deep artificial neural network. CNN has kinds of merits, such as multidimensional data input, and fewer parameters. However, the network always has the problem of overfitting due to lots of connection in the full connection layer. In order to overcome the overfitting problem, the denoising method is used to corrupt input data and hidden unit output which will enforce the network learning a better feature representations of the sample data. In the simulation, some situations are considered, such as input data corruption and hidden unit output corruption, and a comparison is exhibited.

Convolutional neural network is a model of deep neural network, which uses the convolution and sub sampling to realize feature extraction. However, the network is easy to over fitting. In this paper, the denoising method is used to corrupt the sample and force the network to learn the better representations to overcome the over fitting problem. The generalization of the convolutional neural network will be enhanced by this. The simulations exhibit the learning process.

Sparse autoencoder is the fundamental part in some deep architecture. The hidden layer output is the compression of the input data which gives a better representation of the input than the original raw input. However, the determination of hidden unit number is always experiential. In this paper, the different hidden unit number is discussed. The weight of sparse autoencoder will learn the digital number outline of the handwriting instead of pen strokes when the hidden unit number is smaller. The weight can learn the pen strokes of the handwriting when the hidden unit number is larger.

This paper analyzes the design requirements of high-performance EPS controller, on the basis of which, we select Freescale Power Architecture ECU MPC5602P special for EPS. By the means of Modular Design, we propose a new design for EPS controller hardware, which focuses on MPC5602P minimum system, power supply module, three-phase induction motor inverter module and sensor signal conditioning module. Meanwhile, we completed EPS controller program design process in the light of the induction motor vector control theory. Finally, our controller realized the control functions of the electric power steering motor after it was debugged.

Optical waveguide propagation loss measurement method based on optical multiple reflections detection is presented in this paper. By using a precision reflectometer, uncertain influence on waveguide propagation loss measurement caused by fiber-waveguide coupling can be eliminated effectively and the waveguide net propagation loss can be measured accurately. To demonstrate this, the propagation loss of a Silicon-on-Insulator (SOI) rib waveguide fabricated by RIE is measured with the obtained value being 4.3 dB/cm. This method provides a non-destructive means for evaluating waveguide propagation loss. (c) 2005 Elsevier B.V. All rights reserved.

PID controller is used in most of the course-keeping closed-loop control of Unmanned Surface Vehicle (USV). However, the parameters of PID are difficult to tuning. In this paper, we adopt an elitism estimation of distribution algorithm (EEDA) to optimize the PID, which makes use of the probabilistic model to estimate the optimal solution distribution. It has a better global searching ability. A linear Nomoto model is adopted to simulate the USV, and the PID controller is used to control the course of the USV. The simulation results exhibit the validity of the EEDA.

The vessel collision accidents cause a great loss of lives and property. In order to reduce the human fault and greatly improve the safety of marine traffic, collision avoidance strategy optimization is proposed to achieve this. In the paper, a multiobjective optimization algorithm NSGA-II is adopted to search for the optimal collision avoidance strategy considering the safety as well as economy elements of collision avoidance. Ship domain and Arena are used to evaluate the collision risk in the simulation. Based on the optimization, an optimal rudder angle is recommended to navigator for collision avoidance. In the simulation example, a crossing encounter situation is simulated, and the NSGA-II searches for the optimal collision avoidance operation under the Convention on the International Regulations for Preventing Collisions at Sea (COLREGS). The simulation studies exhibit the validity of the method.

Recently, ship collision avoidance has become essential due to the emergence of special vessels like chemical tankers and VLCCs (very large crude carriers), etc. The information needed for safe navigation is obtained by combining electrical equipment with real-time visual information. However, misjudgements and human errors are the major cause of ship collisions according to research data. The decision support system of Collision avoidance is an advantageous facility to make up for this. Collision risk evaluation is one of the most important problems in collision avoidance decision supporting system. A review is presented of different approaches to evaluate the collision risk in maritime transportation. In such a context, the basic concepts and definitions of collision risk and their evaluation are described. The review focuses on three categories of numerical models of collision risk calculation: methods based on traffic flow theory, ship domain and methods based on dCPA and tCPA.

This paper is taking actual control demand of rotary kiln as background and builds a calcining belt state space model using PO-Moesp subspace method. A novel order-delay double parameters error criterion (ODC) is presented to reduce the modeling order. The proposed subspace order identification method takes into account the influence of order and delay on model error criterion simultaneously. For the introduction of the delay factors, the order is reduced dramatically in the system modeling. Also, in the data processing part sliding-window method is adopted for stripping delay factor from historical data. For this, the parameters can be changed flexibly. Some practical problems in industrial kiln process modeling are also solved. Finally, it is applied to an industrial kiln case.

PID controller is used in most of the current-speed closed-loop control of permanent magnet synchronous motors (PMSM) servo system. However,Kp,Ki, andKdof PID are difficult to tune due to the multiple objectives. In order to obtain the optimal PID parameters, we adopt a NSGA-II to optimize the PID parameters in this paper. According to the practical requirement, several objective functions are defined. NSGA-II can search the optimal parameters according to the objective functions with better robustness. This approach provides a more theoretical basis for the optimization of PID parameters than the aggregation function method. The simulation results indicate that the system is valid, and the NSGA-II can obtain the Pareto front of PID parameters.

Optical bistability (BI) and self-pulsation (SP) in high-$Q$ silicon microring resonators (MRRs) induced by thermo-optical (TO) effect and other nonlinear effects are theoretically studied with coupled mode theory and linear stability analysis method. It is found that the boundaries for both BI and SP are mainly restricted by two counteracting effects: free carrier dispersion effect and TO effect. If the refractive index changes of a MRR caused by these two effects are on the same order of magnitude, the output power will exhibit much more complicated dependence on the input power and wavelength, namely, input-power-dependent multi-BI and multi-SP regions will exist at certain input wavelength range. The controllability of multi-BI and multi-SP phenomena by the input power and input wavelength could be very useful in all-optical nonlinear devices.