Table of contents

2019 3rd International Conference on Artificial Intelligence Applications and Technologies (AIAAT 2019) was held in Beijing, China from Aug. 1 to 3, 2019. AIAAT 2019 was organized by Hong Kong Society of Mechanical Engineers (HKSME), sponsored by Xidian University, York Univeristy, Donghua University, Fudan University. The conference provides a useful and wide platform both for display the latest research and for exchange of research results and thoughts in Artificial Intelligence Applications and Technologies and other topics. The participants of the conference were from almost every part of the world, with background of either academia or industry, even well-known enterprise. The success and prosperity of the conference is reflected high level of the papers received.

The proceedings are a compilation of the accepted papers and represent an interesting outcome of the conference. This book covers 6 chapters: system modeling; neural network; computational intelligence; pattern recognition; computer vision; automation and control.

We would like to acknowledge all of those who supported AIAAT 2019. Each individual and institutional help were very important for the success of this conference. Especially we would like to thank the organizing committee for their valuable advices in the organization and helpful peer review of the papers.

We sincerely hope that AIAAT 2019 will be a forum for excellent discussions that will put forward new ideas and promote collaborative researches. We are sure that the proceedings will serve as an important research source of references and the knowledge, which will lead to not only scientific and engineering progress but also other new products and processes.

Prof. Dan Zhang

York University, Canada

List of Conference Chairman, Program Committee, Advisory Committee, Organizing Committee, Local Committee, Technical Program Committee are available in this PDF.

All papers published in this volume of IOP Conference Series: Materials Science and Engineering have been peer reviewed through processes administered by the proceedings Editors. Reviews were conducted by expert referees to the professional and scientific standards expected of a proceedings journal published by IOP Publishing.

In view of the dynamic time variation and multirate data problems in crude furfuryl alcohol distillation process, too few samples of modeling data caused poor generalization ability of soft sensor model and reduced the accuracy of soft sensor model prediction. To solve this problem, this paper proposes a semi-supervised soft sensor modeling method based on cosine similarity-discount(SSSMI-COSD). By calculating the cosine similarity between labeled and unlabeled samples in the same time interval, and combining with the proposed constraints, the data clustering between labeled and unlabeled samples is realized. In addition, in order to avoid the ill-conditioned problem caused by high-dimensional input variables, the clustered data are fused by discount factor value (DFV). An actual data of a crude furfuryl alcohol distillation process (CFADP) simulation experiment were carried out. The results show that the proposed SSSMI-COSD method can effectively improve the soft sensor model prediction accuracy for a crude furfuryl alcohol distillation process.

The monthly import and export data are usually with the challenging characteristics including large scale, nonlinear and hard to fit, leading to their development trends can be hardly predicted. To tackle this problem, we propose to leverage the LSTM-based recurrent neural network to forecast the development trend in this paper. We demonstrate the effectiveness of our approach based on the monthly import and export data of Shandong Province from January 2001 to June 2018. In particular, we achieve the MSE score of 124.39, which outperforms the traditional time series model less than 12.2%.

Current prediction methods on public opinion trend of network news are often carried out according to human experience or traditional time series prediction, but they sometimes lack scientific principle and immediacy. This paper researches on the prediction of public opinion trend of network news based on deep learning. Firstly, we use the BP neural network and LSTM to predict public sentiment trend, then combine word2vec with LSTM to classify the text sentiment and finally use doc2vec algorithm and k-means to cluster the text. Experiment result shows that the mean square error with good performance, which reveals the beneficial effect of our prediction method.

A novel planar translational parallel robotic mechanism with three limbs is designed. The degree of freedom (DOF) and output characteristics of mechanism are studied based on the screw theory. The mathematic models including position, velocity and acceleration are derived according to different forms of the actuated inputs. When the linear displacements of two cylindrical joints are selected as the actuated inputs, the velocity Jacobian is an identity matrix and the mechanism has fully isotropic kinematics characteristics. Singularity of the mechanism is discussed and all singular configurations are explored. Finally, kinematics simulation of the mechanism is carried out by ADMAS and MATLAB software, and the kinematic curves are drawn.

Some research on the in-situ radioactivity measurement in seawater and development of underwater spectrometer using NaI(Tl) scintillation crystal are being carried out for the automatic and continuous monitoring in the marine environment. The gamma spectrometer must be efficiency calibrated for the detection of radionuclides in the seawater. For the problem of traditional experiments, a simulation model of in-situ radioactivity measurement in the seawater was established by using Monte Carlo statistical method. The model simulates the in-situ measurement environment of seawater and the characteristics of the underwater spectrometer developed by ourselves. By calculating the interaction between gamma-ray photons emitted by radionuclides in seawater and various atoms in seawater and the internal structures of the spectrometer, the marine detection efficiency of the underwater spectrometer developed was calculated and verified by the field experiments. The results show that this kind of non-experimental calibration method is effective and feasible for the underwater spectrometer and further research and experiments will continue.

Traffic sign detection is an important part of many systems such as autonomous driving, driver safety and assistance. In this paper, the detection capability of SSD for small targets is analyzed and improved based on ssd_300 model. CSUST Chinese Traffic Sign Detection Benchmark (CCTSDB) dataset is used to train the model for Chinese road traffic conditions. The improved model was compared with ssd_300 model. The experimental results show that the mAP of the improved model on the test dataset achieves 0.85, which is 0.13 higher than ssd_300, and the algorithm can reach real-time detection. The improved model can effectively detect three categories of Chinese traffic signs and has strong robustness against various disturbances.

Many existing researches mostly establish a theoretical car-following model from a perspective of statistical physics, but the vehicle driver is the decision-maker of driving behaviors. When modeling the micro-behavior of the vehicle, the driver's driving intention and decision-making basis need to be considered. In order to consider human factors more in microscopic traffic simulation, this paper uses fuzzy inference technology as the basis, selects the headway distance, ideal speed difference, and reasonable spacing difference as the input variables of the control system, considering the free running and car-following state of the vehicle, constructs fuzzy rules and establishes a car-following model.

Artificial Intelligence's turn of philosophy emerges because of the new crisis in philosophy. Philosophy should attach importance to technical thinking to supplement the deficiency of philosophical thinking. Philosophy's "language turn" and "AI turn" are based on language. The paradigm shift is also that we can define "intelligence", "consciousness" and "subjectivity" in the way of conventionalism. With the rapid development of AI technology, we also need to guide the development of technology with "Aesthetic reason" and promote the harmony between science and technology and art, human and nature.

Blockchain has been emerging as a promising technology that could totally change the landscape of data security in the coming years, particularly for data access over Internet-of-Things and cloud servers. However, blockchain itself, though secured by its protocol, does not identify who owns the data and who uses the data. Other than simply encrypting data into keys, in this paper, we proposed a protocol called Biometric Blockchain (BBC) that explicitly incorporate the biometric cues of individuals to unambiguously identify the creators and users in a blockchain-based system, particularly to address the increasing needs to secure the food logistics, following the recently widely reported incident on wrongly labelled foods that caused the death of a customer on a flight. The advantage of using BBC in the food logistics is clear: it can not only identify if the data or labels are authentic, but also clearly record who is responsible for the secured data or labels. As a result, such a BBC-based solution can great ease the difficulty to control the risks accompanying the food logistics, such as faked foods or wrong gradient labels.

Facial modeling is a key step to model visual effects in special movie effects and computer games. In this paper, a method based on the combination of deep learning and feature extraction is proposed for the modeling of 3D face model. Firstly, the face region is located for the captured face image. And then, the facial feature points are extracted by the landmark algorithm and the Convolutional Neural Network (CNN) is used to classify the facial expressions. Next, a special expression 3D face model is created by the deformation of the standard 3D face model based on the facial expressions classification result. Finally, the 3D face model and the extracted facial feature points are combined to perform personalized adjustment of the 3D model to complete a 3D facial expression animation system. The experimental results show that the proposed method can effectively perform the dynamic 3D face modeling which has high reality.

With the development of society and the improvement of living standards, people's pursuit of their own beautiful appearance is getting higher and higher, especially the passion for beautiful features. Therefore, the demand for dental orthodontic surgery is growing in daily life. However, due to the limitations of oral physiological structure, the orthodontic brackets designed in traditional orthodontic surgery are often not the optimal dental arch shape for patients. Therefore, a dental orthodontic software system that can interactively simulate a patient's oral model is particularly important. In this paper, a software system of dental orthodontic robot which can simulate dental orthodontics was developed by using Blender secondary development technology combined with orthodontics and tooth extraction algorithm.

Aiming at the problem of human motion sequence recognition, algorithm based on feature selection and support vector machine is proposed. Firstly, the feature extraction of human motion sequences is obtained by key frame and human joint angle calculation. Then, based on the Pearson correlation coefficient and CFS evaluation function, the algorithm of relevance feature selection is used to search the optimal feature subset from the original feature set. By reducing the dimension of the feature set, the difficulty of classification recognition is reduced. In the classification process, the support vector machine is used as the classifier to complete the recognition task of the human motion sequence. Through the recognition experiment and the contrast experiment, the effectiveness of the recognition algorithm based on feature selection and support vector machine is proved.

The flight attitude control is the core part of the maneuvering process in air combats. Traditional flight attitude control methods have high computational complexity, low flexibility and poor ability to learn sequential feature. This paper proposes a flight attitude control model based on long short term memory network, which utilizes its special gates structure to memorize historical information, and acquire the variation law of the attitude control variable from the time sequential data including the battlefield situation and flight parameters automatically. Moreover, the basic framework and training methods of the model are also introduced, and the influence caused by various LSTM network parameters is deeply discussed. The experiment results show that the proposed model has better prediction accuracy and convergence performance than the traditional recurrent neural network.

According to the characteristics and business needs of small and medium-sized business management, the paper analyzes and designs the functions and performance of ERP platform, applies Internet of Things and artificial intelligence technology, and builds a dynamic mobile intelligent SME ERP platform through business database, rule database and knowledge database. The platform adopts three-tier B/S architecture, uses MySql as the back-end database, uses Java to encode, implements financial management, supply chain management, manufacturing, human resource management and other functional modules, and finally carries out platform testing. The implementation results show that the intelligent SME ERP greatly reduces the workload of business personnel and improves the efficiency of real-time dynamic management of enterprises.

Short-term series forecasting is one of the essential issues in a variety of tasks, such as traffic flow prediction, stock market tendency analysis, etc. Most current methods based on stable or abundant historical data. In this paper, we proposed a novel model called LA-NN. It takes advantages of both long short-term memory (LSTM) network and autoregressive integrated moving average (ARIMA) by a relation integration of them. So as to deal with the situation of insufficient historical data and sudden abnormal changes in data. A comparison with other representative forecast models validates that the proposed LA-NN network can achieve a better performance.

Pedestrian detection in complex scenes has always been a research difficulty in computer vision. The performance of current methods was seriously degraded when the pedestrian is occluded or the size of pedestrians are too small, etc. In this paper, we propose a novel approach based on location model for the detection of multiple pedestrians, which aims to improve the efficiency of algorithms in complex scenarios. In the model, a fully convolutional neural networks for the classification of pedestrian and non-pedestrian are trained to learn pedestrian features first. Then trained model are used to search for pedestrian regions, and the regions where pedestrian may be present will be activated and marked with boxes. Finally, we fine tune the boxes to overlap them with the ground truth more precisely. Compared with the current methods on two pedestrian datasets, experimental results demonstrate the comparable performance of our approach in term of miss rates.

In order to achieve the on-orbit performance of the satellite, the installation positions of some instruments on the satellite are measured when the satellite is integrated on the ground, to ensure the installation position and the accuracy meet the design requirements. With the development of aerospace industry, more and more satellites have the characteristics of difficult development, short development time and heavy position measurement task. In view of the mass produce of satellites, an automatic angle measurement scheme based on robots is designed and applied. Giving full play to the advantages of automation and high measurement efficiency of robots, this system effectively improves the measurement efficiency and reducing the occupancy of personnel.

Online prediction of subcutaneous glucose concentration plays a critical role in glucose management for type 1 diabetes. In this work, a new method combining Variational Mode Decomposition (VMD) and Least Squares Support Vector Regression (LSSVR) is proposed with three main stages to improve the prediction accuracy. Firstly, the time series of blood glucose are decomposed into different frequency series by VMD method. Secondly, the LSSVR model is trained to predict each subsequence. Finally, the predicted sequences are reconstructed to obtain the overall glucose predictions. The experimental results demonstrate the effectiveness and accuracy of the proposed model for short term glucose prediction.

Three types of dual-rate digital controllers are designed to suppress the output voltage fluctuations caused by load-current changes for a buck DC-DC switching converter running at 300kHz PWM. The slower-rate outer loop is designed using the IA-PIM method, which stabilizes the outer loop at a slow rate of 25kHz. Since the time lag in sensing the disturbance is critical, a faster loop of 300kHz is added based on three different approaches. The resulting dual-rate control laws are implemented on a single 16-bit fixed-point microcomputer with the clock frequency of 140MHz, via 12-bit A/D and D/A converters. Experiments on the fluctuation of the output voltage caused by load-current transients show that, while the analog regulator suppresses the maximum output fluctuation to be 4.7 % of the output voltage, one of the three digita l control methods achieves 4%, without adding ext ra analog components, such as capacitors as in our previous paper.

Photovoltaic arrays are prone to various failures due to long-term work. In order to quickly and accurately diagnose the type of failure of the PV array and implement online monitoring of the PV array, this paper proposes the BP neural network for PV array fault diagnosis, and proposes a network search method when training BP neural network. And the K-cross-validation method is used to select the number of hidden layer nodes. The BP neural network fault diagnosis model designed and trained by this method is proved to have high precision.

People can only obtain incomplete information when needle get into soft tissue. The actual conditions limit monitoring results. The kriging, as a statistical interpolation method, can effectively transform discrete point information into facial continuous information. On the basis of displacement data of pork soft tissue, the correlation parameter setting and optimal semi-variance model selection of kriging were studied to determine the kriging interpolation results. The results show that the accuracy of the kriging model can reach millimetre level, the maximum error is about 0.8mm, the average standard error is 0.35mm, and the calculation is simple. It can help doctors provide references when the soft tissue cannot be directly observed, and predict the deformation within the soft tissue.

In the view point of ecological holism, deep neural networks can be view as a machine cognitive agent with overall multi-level nested structure, the representation learning and problem-solving ability of machine cognitive agent with multi-modal perception function is also the results from the perception and judgment of ambient situation environment. In the paper, a machine cognitive method based on ecological holism was deconstructed. Then, two core questions to characterize learnability of AlphaGo were answered, one is how to explain the characteristics of the hierarchical gradient and update parameters between different layers of networks? Two is how to understand the effective ability to choose the drop sampling at the strongest level of game of Go. These core controlling structural and functional roles proposed in this study may be further improved to a new type of artificial intelligence law of machine cognitive system.

Sentiment classification aims to classify the sentimental polarities of given texts. Lexicon-based approaches utilize lexical resources to explore the opinions according to some specific rules, whose effectiveness strongly depends on the goodness of the lexical resources and the rules. Traditional machine-learning methods tightly rely on feature engineering and external NLP toolkits with unavoidable errors. Deep learning models strongly rely on a large amount of labelled data to train their numerous parameters, which often suffer from overfitting issue since it is difficult to obtain sufficient training data. To address the issues, we design a model that combines Knowledge-oriented Convolutional Neural Network (K-CNN) and bidirectional Gated Recurrent Neural Network (biGRU) in a hierarchical way for sentiment classification. Firstly K-CNN is used to capture the n-gram features in sentences. Sentiment word filters are constructed in the knowledge-oriented channel of K-CNN based on the linguistic knowledge from SentiWv ordNet, which can capture the sentiment lexicons and alleviate overfitting effectively. Then biGRU with attention mechanism is utilized to model the sequential relations between sentences and obtain the document-level representation based on the relevance of each sentence to the final sentiment classification. Experiments on two datasets show that our model outperforms other classical deep neural network models.

Financial time series prediction is one of the most complex and challenging problems in both AI and finance engineering. In our research, we proposed a Hybrid Chaotic Oscillatory Neural Network (HCONN) model by replacing the traditional sigmoid-based activation function with chaotic oscillatory activation function, which provides significant performance in the global minimum convergence through the application of Adaptive Moment Estimation optimizer. In addition, by integrating the latest R&D on Quantum Finance Theory (QFT) and its Quantum Price Level (QPL) as the deep features' extraction, we add the daily 8 nearest QPLs together with the time series price variables as the input of our HCONN. In terms of system implementation, 12 different forex products including the AUDCHF, AUDUSD, CADCHF, EURAUD, EURCHF, EURGBP, EURUSD, GBPAUD, GBPCAD, GBPUSD, USDCAD and USDCHF are used. System performance results reveal that HCONN outperforms other financial models including: Feedforward Backpropagation Neural Network (FFBPN) and Chaotic Oscillatory Neural Network (CONN) in terms of training performance and forecast accuracy.

Semantic segmentation is an essential step to do further image analysis and scene understanding tasks. In medical imaging analysis applications, it is even more challenging to do automatic segmentation due to tissues' complicated boundaries. In this paper, a fully convolutional network (FCN) based model is constructed to segment distal radius and ulna (DRU) areas from hand X-ray images. We evaluated the proposed network on a clinical DRU dataset with different network configurations. The proposed network can achieve 98% accuracy and 96% mean Intersection over Union (IoU).

Mining and analyzing online travel reviews and travel information is playing an increasingly important role in the tourism industry. Accurately capturing the uniqueness and attractiveness of the tourist destinations recorded in the travel notes is the key to tourism analysis and application. The current way to obtain the attraction of tourism is easy to cause bias due to the use of simple statistical methods. This paper proposes a model based on deep learning, which uses Bert pre-training method, based on Transformer, and mines travel notes through Attention to find the attraction point. The model can understand the chapter-level semantics of travel notes based on the context, so much so that the extracted features are closer to the meaning of the text. It also exhibits good performance in generating unique labels of tourist destinations and similar tourism clusters. The experimental results are consistent with the facts, the validity of the model is also proved.

Under the background of the rapid development of Internet technology and the popularity of smart grids, the analysis and prediction of short-term time series data of users' power consumption has important guiding significance for grid planning, management decision of economic sector and optimization and allocation of power resources. Considering that the traditional statistical-based time series analysis method is weak in generality and can not handle the complex linear problem in prediction, the long-term dependence of the ordinary cyclic neural network model is insufficient, and the time series data has multidimensional problems, a deep neural network is proposed. The PCA-LSTM model is used for time series data prediction. The model firstly uses the PCA (principal component analysis) method to reduce the dimensionality of the electricity consumption time series data, optimizes the number of input variables, and inputs the data into the long- and short-term memory network LSTM for training prediction. The experimental results show that the LSTM network prediction based on PCA improves the accuracy of short-term time series data prediction, and also improves the convergence speed of LSTM network. It proves that the method has better prediction performance and versatility.

Human parsing is still a big challenge in computer vision task to accurately label every body part in an image. Current methods of sematic segmentation mainly focus on dividing each independent part and ignore the structural priors of human body. However, the position relationships of different parts, especially which around the area of boundary, indicate their associations and also contribute to the semantic segmentation. In our work, we propose a body-boundary-refined part to refine the segmentation result of human part edge by simply utilizing the structure priors around the body boundary. It puts a penalty mechanism on wrong marginal pixels to improve segmentation performance around the area of body boundary. The network achieves competitive performance on the PASCAL-Parts-dataset and especially the area around the body boundary has been refined.

On the background of "Internet +" era, the domestic higher education is showing the trend of artificial intelligence. The reliability and scientificity of computer intelligent evaluation are further carried out, and the mode of intelligent evaluation and data analysis in optimizing the precise teaching of English writing is explored, which can lay a foundation for the large-scale use of the technology. Based on data-driven theory, this study further analyzed the role of AI in promoting in-depth learning by comparing AI writing review model with manual review model.

For expensive black-box problems, surrogate modelling techniques are generally used to decrease the computational source. In this study, an improved surrogate based optimization (SBO) method is presented to solve the real-world engineering applications with expensive black-box objective responses. An optimized ensemble of surrogates combing three typical surrogate modelling techniques is adapted to efficiently predict the objective response. Meanwhile, the hierarchical design space reduction (HSR) strategy is employed for obtaining the smaller design subspace for improving the optimization efficiency. During the search, all test problems are considered as the real-world engineering applications whereas the actual global optima as well as the function characteristics are unknown in advance. The results show that the proposed method is superior in identifying the global optimum.

In the process of pipeline transportation, the temperature and pressure of waxy and high viscosity crude oil are decreasing. In order to complete the pipeline transportation task, stations are usually set at intervals to overcome the pressure loss caused by friction and impact in the flow and the temperature loss caused by the radial temperature difference. The main equipment of the station is pumps and heaters. The operating fee of pumps and heaters is the core costs of hot oil pipelines, so it is necessary to reasonably optimize the pump and heater operating schemes. Because the hydraulic and thermodynamic of the hot oil pipeline are coupled with each other, and there are a lot of nonlinear constraints and integer variables, this optimization problem is a kind of mixed integer nonlinear programming, which is difficult to solve by using the traditional optimization method. In this paper, a hybrid meta-heuristic algorithm based on particle swarm and pattern search is proposed, which can deal with the problem effectively. The operational cost of the optimized scheme is much lower than that of the actual operating scheme. So this method can provide guidance for the cost reduction and efficiency improvement of the hot oil pipeline.

The basic genetic algorithm has the disadvantages of falling into local optimum and slow convergence. To solve this problem, a hybrid algorithm combining simulated annealing strategy is proposed. The cooling process in simulated annealing is used to complete the iterative process in the hybrid algorithm. The algorithm is used to solve the traveling salesman problem. The results show that the convergence speed and accuracy of the hybrid algorithm is significantly better than the basic genetic algorithm.

In this paper, we improve the effect of perimeter control for Urban Traffic Sub-region by optimizing signal timing in traffic sub-region. Firstly, we verify the influence of different signal timing on the shape of macroscopic fundamental diagram(MFD) by simulation. Then, we use BP neural network to model the mapping relationship between signal timing and MFD maximum volume(saturated traffic volume of sub-region). In perimeter control based on MFD, the larger the saturated traffic volume, the more vehicles passing through the sub-area. When the output of the BP neural network model reaches the maximum value, the signal timing of the road network is the best one. Comparing the perimeter control effects of different signal timings, we find that the larger the saturation traffic, the larger the traffic in the boundary control and the smaller the vehicle delay.

Laser scanning system provides an efficient solution to rapidly acquire 3D information of large-scale scenes. Point clouds collected by laser scanning systems contain numerous objects with significant disparities in size, complicated and incomplete structures, holes, varied point densities, and huge data volumes, raising great challenges for automated point clouds registration, segmentation, and object detection. The dissertation presents a hierarchical merging based multi-platform point clouds registration algorithm to align MLS point clouds and unordered TLS point clouds from various scenes and validates its performance on nine challenging datasets. The algorithm improves the efficiency and accuracy of point cloud registration and enhances the registration ability of the algorithm for low-overlap and high-symmetry point clouds.

The crankshaft is a high-speed rotating component that is widely used in many fields. Dynamic balance of the crankshaft is an important process stage that affects the quality of the crankshaft. In order to optimize the quality of crankshaft dynamic balance machining, this paper illustrates the importance of the initial unbalance of crankshaft to the dynamic balance machining of crankshaft. For the initial imbalance distribution of crankshaft, the improved mean method, gradient descent method and mean shift algorithm are used to calculate the machining center. The position of the hole realizes the crankshaft imbalance correction with different effects and reduces the processing strength of the crankshaft dynamic balance phase. The study is applicable to the machining of crankshafts in geometric centering processes in batches.

Based on the information management needs of SMEs, the article adopts the rapid prototyping method, designs the overall framework of the cloud ERP system, maps out the functional modules, adopts the B/S architecture, uses Java as the development tool, and SQL Server as the background server. The cloud computing and APP technologies are used to realize instant access between the PC and the mobile terminal, and the cloud ERP system with relatively complete system functions, simple interface and mobile operation is realized. The system incorporates new technologies such as big data analytics, artificial intelligence, machine learning and data mining to enable it to realize intelligent financial analysis, intelligent financial forecasting and intelligent financial decisions.

The distributed permutation flowshop scheduling problem (DPFSP) has attracted many researchers' attention in recent years. In this paper, we extend the DPFSP by considering the sequence-dependent setup time (SDST). A new hybrid genetic algorithm (HGA) for the DPFSP with the SDST (SDST/DPFSP) is presented to minimize the maximum of the completion time. At first, a new population initialization is proposed. And then, the newly-designed operators are described in details, and we also introduce the mutation rate and the crossover rate to balance the mutation operator and the crossover operator. To further improve the obtained solution, a new local search method is developed. At last, the orthogonal experimental design is applied to adjust the parameters in the HGA, and a comprehensive computational campaign based on the 135 instances demonstrates the effectiveness of the proposed HGA for the SDST/DPFSP.

In the multiple unmanned aerial vehicle (multi-uav) cooperative tracking target control system, the control system is extremely complex due to the nonlinear control model and constraints. In this paper, multi-uav cooperative tracking target motion modeling is adopted, optimization performance indicators is established, and multi-objective rolling optimization is transformed to single-objective optimization. Finally, particle swarm optimization (PSO) prediction control algorithm is adopted for optimization solution, realizing real-time control of multi-uav cooperative tracking target, and its feasibility is verified by simulation.

Higher vocational college students' entrepreneurial ability is based on the characteristics of higher vocational students and that of vocational colleges. It has similarities in entrepreneurial ability, but also has their own characteristics. Different from emphasis on scientific and technological entrepreneurship, higher vocational college students focus on the fields such as service and sales, and the small and micro entrepreneurial ability of higher vocational students emphasizes the use of skills and opportunity development. This paper makes an empirical study on the small and micro entrepreneurial ability of higher vocational students, and finds that their entrepreneurial ability is composed of skills practice ability, opportunity development ability, entrepreneurial operation ability and failure handling ability. The small and micro entrepreneurship of higher vocational students is an important part of college students' entrepreneurship. The development of their ability will improve the employment competitiveness of higher vocational students and the success rate of entrepreneurship.

Bio-inspired computation aims to study the biology function, characteristic and echanism of the various levels of nature from biological individual, population, colony until ecosystem, and set up a relevant model and computing method, so as to serve the scientific research and engineering application of human society. A few future directions and research challenges are presented, such as parallel bio-in-spired computation, bio-inspired computation with reasoning and knowledge, bio-inspired dynamics computation, bio-inspired computation based on quorumsensing, artificialbrain, evolutionaryhardware, bigdata, swarmrobot, virtualbiological, cloudcomputing, etc.

The combustion product of solid propellant produces strong plume flow field downstream of nozzle outlet, which improves plume characteristics, reduces plume characteristic signals, and improves stealth characteristics and survivability of missile. In this paper, an intelligent algorithm is proposed to optimize the formulation of solid propellant with low characteristic signal. An improved multi-objective particle swarm optimization algorithm is used to solve the formulation optimization model, and several feasible formulations for realizing low characteristic signal are obtained. On this basis, considering the energy performance, a formula screening method is given to calculate various characteristic signals and energy properties of the feasible formulation. The comprehensive measurement can not only take into account the optimization target requirements of propellant energy performance and plume characteristic signals, solve the conflict problem of reducing multiple characteristic signals, but also improve the efficiency of formulation design, and provide a reference for formulation designers.

Multi-Agent Path Finding (MAPF) problem is intensively studied in theoretical computer science, robotics and so on. The key for MAPF problem is to plan a conflict-free path for each agent with different start and goal positions, and to minimize the cost of the paths. Conflict-Based Search (CBS) is one of the optimal algorithms, which can ensure that the optimal solution is obtained, in the case of small-scale maps and low number of agents. However, in many real-world multi-agent systems, the scale of the map is very large and the number of agents is more. In most cases, CBS is not applicable any more. Therefore, we develop Divide and Conquer CBS, called DC-CBS algorithm which divides large-scale map and the original problem into several smaller ones. For each subproblem, we use the CBS to get the optimal solution. The experimental results show that for the large-scale scenario, the DC-CBS algorithm can get the result more quickly and plan the paths for agents successfully. As for small-scale situation, the performance of DC-CBS is almost as good as that for CBS.

Clothing knowledge graph is a kind of vertical domain knowledge base constructed for the description of clothing knowledge in the field of textile and apparel. In this paper, based on the limitations of the clothing knowledge graph in the effect of entity extraction, the deep learning model and the statistical model are combined. A Chinese named entity recognition method based on CNN-BiLSTM-CRF is proposed. Firstly, the convolutional neural network(CNN) is used to extract the text features, and the character-level vectors with morphological features of the words are trained. Then the bi-directional long short term memory networks(LSTM) is used to learn the context features, and the vector representation of the context of each word is output. Finally, the conditional random fields(CRF) model is used for self-learning. Get the best tag sequence for the sentence. The method can automatically recognize the text, and does not rely on the artificial feature to obtain the semantic category information. Finally, the experimental data and evaluation methods are introduced. The experimental results show that the Chinese named entity recognition method based on CNN-BiLSTM-CRF is superior to other models in all indicators, indicating the effectiveness of the method.

Style crack refers to the position where the author's identity changes in the article completed by multiple authors. This paper summarizes the current situation and theory of related fields at home and abroad, and proposes a multi-feature based document segmentation method for plagiarism detection. Seven text style features are used for style crack recognition. Through the result of feature extraction, the combination of multi-feature fusion and unsupervised machine learning algorithm is used to classify the features based on extraction, and the clustering algorithm is used to cluster the style features so as to find the location of style cracks. Experiments show that the method is effective and scientific, and achieves good results.

Reserve capacity optimization of power system plays an important role in large-scale Wind Power System. In this paper, a novel multi-information fusion algorithm based on population (MIFA-P) is proposed, which can balance the problem of reserve capacity optimization. MIFA-P determines the type of optimization problem by counting the maximum number of projections on the principal characteristic axis of the optimal individuals of the current population. Therefore, the corresponding class of evolutionary algorithm is chosen to solve the problem. In the process of evolution, each evolutionary algorithm achieves the organic integration through sharing population information. Finally, the validity and applicability of MIFA-P algorithm for large-scale decision variable black box optimization problem are verified by solving a practical scenario optimization problem of reserve capacity distribution in power system after wind power integration.

The collection of Chinese-Myanmar Parallel Corpus (CMPC) is the key step in the natural language processing (NLP) and training Machine Translation Engine (MTE) of Southeast Asia minority languages. As the scarcity of CMPC resources that efficient corpus collection methods are worth studying extremely. Traditional corpus collection methods include manual collection, text recognition of books and Internet crawlers, etc. Among them, the most efficient method to collect corpus is internet crawler preached by many. Traditional Internet crawler algorithm is interfere easily by a lot of spamming and advertising that lead to the time-consuming and low-precision. We propose a web crawler mechanism combines acquisition automatically technology bilingual website list, crawling corpus and cleaning corpus to obtain high quality parallel corpus. Firstly, using the hyperlinks to recursively access related corpus websites through building the website graph. Furthermore, the breadth-first, Backline and PageRank crawler framework used to build a corpus selection model based on crawling with threshold, matching link, ranking the heat of page, through this, the CMPC can be found accurately. Finally, the corpus cleaning model based on the HTML parsing to determine a set of standardized token sequences. By testing the Chinese-Myanmar reptile algorithm established in this paper, the experimental results show that our benchmarks this model exceeds previous published benchmarks. Up to now, we have obtained 1.1 million parallel corpus pairs of Chinese-Myanmar.

Positron Emission Tomography (PET) allows tumour microenvironment to be studied in vivo with high sensitivity and specificity. Inter- and intra-tumour morphological and phenotypic heterogeneity or pattern provided by PET images are of critical importance. The traditional practice of visual interpretation of these images are not sufficient enough to extract all the information embedded in the images. On the other hand, simultaneous development of automated and reproducible analysis methodologies makes it possible to extract large amount of quantitative features from these images which is termed as radiomics. Analysis of these radiomics feature using artificial intelligence (AI) can significantly improve individualized treatment selection and monitoring. Grey level co-occurrence matrix (GLCM), a member of texture based radiomics feature family is widely used as a biomarker of heterogeneity and can provide information of the tumour microenvironment. The GLCM can subsequently be used for artificial intelligence (AI) assisted tumour diagnosis, monitoring of progression and treatment planning as well as for monitoring response to therapeutic intervention. This aim of the study was to investigate the accuracy and robustness of PET based GLCM in varying image acquisition and analysis conditions using phantom data. It has been observed that GLCM based textural features (e.g., correlation, entropy, homogeneity, energy contrast and dissimilarity) are not only dependent on the volume but also on the quantization level. They are also dependent on signal-to-noise ratio (SNR) and image contrast. The dependencies of these features to the varying imaging conditions are also not linear and cannot always be directly related. To use these GLCM derived textural features as biomarkers for AI assisted analysis, all the information regarding the textural features should always be included along with the changes in volumes and contrast of the PET images in the training dataset.

Automated in-vitro cell detection and counting have been a key theme for artificial and intelligent biological analysis such as biopsy, drug analysis and decease diagnosis. Along with the rapid development of microfluidics and lab-on-chip technologies, in-vitro live cell analysis has been one of the critical tasks for both research and industry communities. However, it is a great challenge to obtain and then predict the precise information of live cells from numerous microscopic videos and images. In this paper, we investigated in-vitro detection of white blood cells using deep neural networks, and discussed how state-of-the-art machine learning techniques could fulfil the needs of medical diagnosis. The approach we used in this study was based on Faster Region-based Convolutional Neural Networks (Faster RCNNs), and a transfer learning process was applied to apply this technique to the microscopic detection of blood cells. Our experimental results demonstrated that fast and efficient analysis of blood cells via automated microscopic imaging can achieve much better accuracy and faster speed than the conventionally applied methods, implying a promising future of this technology to be applied to the microfluidic point-of-care medical devices.

A novel ground plane detection method is put forward based on three-dimensional (3D) point cloud data obtained using an RGB-D sensor. It consists of three stages: data pre-processing, occupancy grid map construction and ground plane segmentation. In order to obtain more accurate 3D point cloud data, the weight median filter (WM) is applied to recover the invalid depth pixel in the depth image. Different from the traditional approaches which process the 3D point cloud data directly, our algorithm transforms the 3D point cloud data into an occupancy grid map. Considering that the occupancy in the occupancy grid map and the distance from the point to the ground plane are distinct between ground points and other points, we will get a part of the 3D points that is definitely in the ground. The points selected are back-projected to the pixel coordinate system to get the result of the ground plane detection. In terms of experiment, the sensor is mounted on the mobile robot, Turtlebot2. The proposed method can detect more than 95 percent of the ground point. It produces accurate ground plane detection in different scenes.

Depth and visual odometry estimation are two essential parts in SLAM systems. Compared with traditional algorithms, supervised learning methods have shown promising results in single view depth estimation and visual odometry estimation. However, they require large amounts of labeled data. Recently, some unsupervised approaches to estimate depth and odometry via minimizing photometric error draw great attention. In this paper, we present a novel approach to learn depth and odometry via unsupervised learning. Our method ameliorates the original photometric loss to enhance the robustness to illumination change in real scenarios. In addition, we propose a new structure of Pose-net and Explainability-net to achieve rotation-sensitive odometry results and more accurate explainability masks. The experimental results have demonstrated that our approach achieves better performance than existing unsupervised methods in both depth and odometry results.

The Person re-ID task objective is to search for a specified target pedestrian image in a large gallery. With the development of deep learning technology, the accuracy of personnel re-identification has been greatly improved. However, previous methods lacked sufficient attention to local features when focusing on global features, causing background and occlusion objects to affect the foreground. The framework we proposed specifically designed this problem to optimize local features through human pose key-points and self-attention mechanisms. And the effectiveness of the method is demonstrated in multiple datasets.

Terrain recognition is one of the key problems of mobile robots. It can help the robots understand the surrounding environment. With terrain prediction, the robots could realize autonomous navigation and path planning. This paper focuses on image feature selection for terrain recognition with visions. For terrain recognition tasks, feature is used to represent image information. Traditional visual features can be targeted to express the low-level information like color or texture. The deep feature is extracted by self-learning of neural network, containing richer semantic information than low-level features. There is a complementary relationship between the two. The efficiency and accuracy of terrain recognition is remarkably raised by the fusion of two above features. In the course of algorithm, combination of the off-line training model and on-line recognition model is used to identify the terrain type of the sample, which is to ensure the real-time performance. The corresponding terrain dataset--SDUterrain is established. The algorithm achieves 96% or higher classification accuracy in the experiments based on the SDUterrain Dataset, which is much higher than the single feature classification algorithm.

This paper treats person re-identification (re-id) as a sequential model, guide person re-id with person detection, combines recurrent neural network (RNN) with attention mechanism, and proposed an end to end person re-id method for surveillance scenarios. The feature of target person is firstly extracted using ResNet, and then the feature is added to the long short-term memory (LSTM) network to guide the attention model for the region of interest in the surveillance image. Finally, combined with the information observed from the image multiple times, the most similar candidate person in the image is deduced, and the feature distance is calculated and ranked for person re-id. This paper strengthens the relationship between person detection and person re-id, and reduces the error between models. Because the number of candidate person matched with target person is reduced, this method can process person re-id task with less calculation and time. This paper also verified the effectiveness of the proposed method by experiments comparing a variety of person detection and re-id methods on several person re-id datasets.

In order to improve the recognition rate of real-time classification of facial expressions, we proposed a method of facial expression recognition based on voting mechanism. Firstly, different neural network models are constructed to learn facial features. Then, the extracted features are fed into the classifier to obtain the posterior probability of various features. Finally, through the voting mechanism, the optimal decision-making level fusion is achieved to complete the facial expression classification. Experiments show that the average recognition rate of fer2013, CK+ and JAFFE database is 74.58%, 100% and 100% respectively. Compared with other recognition methods, experiental data show that this method has superior performance, improves the recognition rate and robustness of the algorithm, and ensures the universality of the algorithm.

Intrusion target detection and recognition are of great significance to security protection of oil and gas fields. An intrusion detection system is built with the integration of infrared image acquisition module, infrared image processing module, moving target detection module and recognition module. Traditional target recognition algorithm highly relies on manual design feature extraction algorithm, which requires designer to have adequate prior knowledge, and cannot avoid the influence of subjective factors of people. Intrusion detection and target recognition system are proposed based on deep learning, which uses neural network algorithm. Deep learning model is built through feature extraction and training of acquired images of intrusion objects, and thus subsequent invasion objects are detected and recognized. Intrusion detection is achieved through simulation of human brain, which boasts of more intelligent recognition process and more accurate recognition results compared with traditional recognition method. According to applications in real scenario, the system proposed has better detection and recognition results and great practical value.

In this paper, we introduce the Extracting specific human 3D skeleton point system based on monocular tracking. The system mainly consists of two parts. The first part is the detection and tracking of specific human body. This article uses simple online and real time tracking with a deep association metric (DEEP SORT)[1] algorithm, which is simple but effective, and meets system requirements in terms of efficiency and real-time. The second part is to extract the 3D bone points for the specific target of the tracking. We refer to Xingyi Zhou's research work[2] in this area. Utilizing the correlation between 2D pose and depth estimation subtasks, the training is end-to-end, and the algorithm introduces 3D geometric constraints to normalize 3D pose prediction, which is effective without ground truth value depth labels. In this paper, the two methods are combined by improvement, and the Extracting specific human 3D bone point system based on monocular tracking is designed. It can realize the tracking of 3D skeletal points of specific targets. The system has high practical value in human-computer interaction, virtual reality and motion recognition.

Action recognition technology is an important part of artificial intelligence. In order to improve the rate of action recognition, this paper designs an action recognition method based on Weighted DTW algorithm. Firstly, the Kinect2.0 is used to obtain the three-dimensional data of the human joint points. Then the quaternion method is used to define the action sequences. The weight of joint is calculated according to the participation in different types of action. Then the improved DTW algorithm is used to design the action recognition experiment. The experimental results show that the action recognition algorithm designed in this paper has better recognition rate and timeliness than traditional DTW algorithm and F-DTW algorithm.

The multi-agent technology is used to reasonably allocate the search area determined after the aircraft crash, and the research on the task assignment problem of aircraft crash area based on multi-agent system is proposed. In the simulation experiment, the model of multiple rescue points, single disaster points and multiple emergency resource allocations is shown. The results show that the method can optimize the allocation of the search task in the aircraft crash area and dynamically change with the change of the crash area. The aircraft's search-assisted decision-making system provides some support.

For the spin stabilized two-dimensional trajectory correction projectiles with utation and precession phenomenon, an attitude controller with sliding mode control method and unidiational auxilairy surfaces (UAS-SMC) is proposed in this paper. And the underactuated control allocation in two-dimensional trajectory correction projectile with UAS-SMC controller is also discussed here. Simulation results show that the proposed controller can effectively control the attitude of the projectile and has strong robustness, which provides reference for engineering applications.

For the traditional calibration method, the problem of zero offset and amplitude variation caused by degaussing during the launching of the rotating projectile cannot be solved. Based on the practical engineering application, an automatic initial calibration method for the geomagnetic sensor combination is proposed. Based on the spherical harm onic model of the Earth's magnetic field, the output law of the non-orthogonal geomagnetic sensor combination and the instantaneous attitude of the projectile are analyzed, and the geomagnetic sensor combination is automatically calibrated. The experimental results show that after the automatic calibration in the air, the geomagnetic data error collected by the experiment is significantly reduced.

Circular trajectory is most frequently used when an autonomous underwater vehicle calibrates positions of underwater acoustic beacons or corrects its own navigation errors using beacons. However, the space-time effect on control and navigation data processing and the impact of the angle of attack on three-dimensional localization performance of AUVs have rarely been mentioned in previous studies. To improve the three dimensional localization performance of AUVs with a single-beacon aiding, methods for space-time trade-offs from a control perspective and determination of the optimal angle of attack are investigated in this study. Through theoretical deduction aided with numerical simulation, concerned quantitative values with their corresponding preconditions are given, and some general conclusions which can greatly improve the navigation capability of AUVs have been made.

The research of Unmanned Aerial Vehicle (UAV) flight autonomous landing on a moving platform based on computer vision has important significance, especially for the rescue and spray UAV in the future. A vision-based relative position and attitude estimation algorithm with a novel color marker and its recognition method was proposed for the UAV autonomous landing system. The new marker is designed with a pattern which make it is as visible as possible in the whole landing process. The proposed relative position and attitude estimation algorithm is based on the four corner points on the marker which are detected by the combined approach of vanishing point of parallel lines and Levenberg-Marquardt (L-M) optimization method. Indoor and outdoor experiments were carried out. The results indicate that the method of marker detection has real-time and robust performance and the position and attitude estimation accuracy attain cm-level, which shows the feasibility of landings on moving platform automatically.

In order to determine the movement trajectory and characteristics of puncture needle in tissue, a series of puncture is selected for the puncture of PVA prosthesis. The image captured by CCD camera is preprocessed, and the starting coordinate point and target in the image are extracted. A fitting model based on least squares method is proposed to fit the radius of puncture needle. Comparing the simulation trajectory of forward kinematics and experimental trajectory, the experimental results show that the fitting model has a high success rate over 93% in the trajectory radius of the prosthesis puncture needle, and has a high success rate, stability and accuracy. It has a certain reference significance for the future path planning and guidance process.

Robotic navigation via speech guidance gain potential attention on the field artificial intelligent and it stat-of-art. This paper highlights speech guide for robot toward machine vision were designed. The robot navigation can determine and identify the objects around the blind area. The process of determining the direction and distance of the blind people to consider for deciding and predicating the next step. The information of the sensors is fused together, and a path information suitable for the blind person is transmitted to the user via speech voice broadcast are the key elements for identification part. Moreover, deep learning algorithm are involve for accurate classifiers target into categories, to archive the highest recognition accuracy. The direction and distance are judgment part to adopt include the improvements binocular ranging, angle measurement algorithm with a consideration of errors marge less than 2%. The novelsystem can-authcorrect and, guide blind people to move froward, advance, stop, turn, which greatly compensates for the visual defects of the blind, with respect of constraint of improving the indoor safety of the blind people.

The speech signal can be modelled as AR models with an innovation noise model. The Pearson type VII distribution is used to model the real excitation. Variational Bayesian framework is used to estimate the posteriors of the AR coefficients and noise model parameters. The model is not conjugate, so MCMC is embed into VB framework to estimate the degree of freedom (DOF) parameter of Pearson type VII distribution. The model order selection is carried out by setting ARD priors on the coefficients. Simulation is carried out on synthetic and real data, the results show that the algorithm performs well for linear prediction both for synthetic data and speech signal, and the results are better than using least square method.

SVPWM algorithm is widely used in the field of power electronic frequency conversion power supply, motor frequency conversion control, power quality optimization and other technical fields. In this paper, the principle of SVPWM modulation is analyzed, in order to make the over-modulation algorithm more simplistic, the adverse effects on the system are smaller, and a phase and amplitude are used to change the modulation method to optimize the algorithm of the cross modulation II zone[1,2]. And using Matlab software to build SVPWM model for simulation verification. Explains the correctness of this method.

A method is introduced to predict human motion trajectory in the process of human-robot collaboration (HRC). In the method, the human-robot distances are assumed to be a Gaussian Process (GP). To achieve this, a human-robot handover task is conducted by a human and a collaborative robot, while the positions of the human hand and the robot end-effector are recorded. Some of the recorded data are used for the Gaussian Process Regression (GPR), a GP and a 95% confidence convince about the GP are obtained by the GPR. Experimental results show that about 80% of the testing data are included in the 95% confidence convince. The method and results here are useful to other human-robot collaborative tasks where existing human-robot relative motions, especially, the method is able to predict the human motion trajectory with varying initial position of the human hand and varying locations of the robot end-effector.

Please see Article PDF for corrigendum text.