Advances in Artificial Intelligence and Security

With the advancement of technology, the analysis of crowd abnormal behavior has become a hot topic in the field of computer vision. The research not only includes the analysis of the abnormal behavior of a single pedestrian in a simple scene, but also includes the analysis of the overall abnormal behavior of the crowd in a complex scene. This paper research on the detection and alarm of abnormal crowd behavior in surveillance video. First, the moving target is detected by the background subtraction method. Secondly, the fall behavior in the video is detected through two-level SVM and human feature action recognition. At this stage, the human body features obtained in the target detection stage are analyzed, and the key point features of the human body are analyzed to determine whether the detection target has fallen behavior. Finally, a counting module is introduced to count the pedestrians on the basis of target detection in the current scene. We compare the changes in the number of people in the video frames at a set time interval in the same scene to determine whether a sudden crowd gathering occurred abnormal behavior.

Image Data are critical data for computer vision. Above all, occluded Image Data is still challenging. This paper, introduces Point Mask, a new data augmentation method for training the convolutional neural network (CNN). In training, Point Mask selects some corners region in an image and erases its pixels with zero values. In this process, training images with various places of occlusion are generated, which reduces the risk of over-fitting and makes the model robust to occlusion. Point Mask is extremely easy to implement and can be integrated with most CNN-based recognition models. Point Mask is very effective and consistently improves over solid baselines in image classification, object detection, and person re-identification. The extensive experiments have demonstrated the new method's the generality and effectiveness.

The existing intelligent crack detection methods of buildings can not meet the needs of crack detection because the surface of buildings has the characteristics of multi-texture and multi-target after earthquake disaster and is interfered by other factors. A three-dimensional measurement technique based on structured light is proposed to detect cracks after earthquake in order to improve the detection rate. Firstly, a raster fringe image with a four-step phase shift is generated and calibrated by a projector connected to a computer. Then, the generated sinusoidal grating fringe image is projected onto the measured object and background, and the modulated grating fringe image of seismic crack surface is collected by a camera and transmitted to a computer. Next the algorithm combining four-step phase shift method and multi-frequency heterodyne method is used to carry out phase unwrapping, and the phase value and height conversion formula are combined to calculate the height to reconstruct the three-dimensional information of cracks generated in the earthquake. Finally, ICP algorithm is used to mosaic images, and the cracks generated after the earthquake are detected combined with the height difference. The experimental results show that the four-step phase shift method combined with multi-frequency heterodyne method is selected to achieve better phase unwrapping accuracy than other algorithms, and the reconstructed error is less than other algorithms. The method based on structured light can accurately detect the spliced fracture image without interference from other factors, which provides technical support for earthquake disaster investigation.

In the task of Fine-Grained Image Recognition (FGIR), the overall difference between different types of images is slight, so locating the representative local region in the image is the key to improving the classification accuracy. This idea of FGIR has been widely used in previous work, and has achieved good results on the benchmark dataset. Recently, the proposal of the Vision Transformer (ViT) method, provides a new method for the field of computer vision. Compared with the previous work based on Convolutional Neural Network (CNN), it has achieved better performance. ViT performs well in general image recognition tasks. However, when applied to FGIR tasks, it only pays attention to the global information and does not pay enough attention to the local features with discrimination. In order to make the model pay more attention to differentiated local regions, we propose an attention-based local region merging method Group Attention Transformer (GA-Trans), which evaluates the importance of each patch by using the self-attention weight inside the Transformer, and then aggregates adjacent high weight attention blocks into groups, then randomly select groups for image crop and drop. Through the weight sharing encoder, the global and local regions of the image are classified after obtaining the features respectively, which is convenient to realize the end-to-end training. Comprehensive experiments show that GA-Trans can achieve state-of-the-art performance on multiple benchmark datasets.

With the development and popularity of the Internet, a huge amount of text data has emerged on the web, and how to accurately and quickly identify the categories of these text data has become a challenge in the field of text classification. To address this challenge, this paper aims to apply deep learning techniques to text classification and build a model that can classify text data in a batch manner and automatically. By studying the current state of the art in text classification needs, this paper proposes a TextGCN model, a text classification method that presents high robustness on small data sets, based on graph convolutional neural networks. This paper makes the model perform better by modifying the IDF formula.

The medical dispute cases are professional and closely related to medicine. Therefore, the mediation of cases in practice depends heavily on similar historical cases. Multi-label classification of legal documents can efficiently filter irrelevant historical cases, which helps to recommend the similar historical cases faster and better. However, the imbalance and label symbiosis of the data set directly affect multi-label classification of legal documents. Therefore, a multi-label classification scheme based on resampling and ensemble learning is presented in this paper The scheme includes two parts: in the first part, in order to reduce the impact of label symbiosis on resampling, a resampling algorithm based on the average sparsity of the label set is proposed improve the imbalance of the data set; in the second one, a multi-label classification algorithm based on ensemble learning is proposed to train multiple base classifiers and combine each base classifier with a voting strategy of one vote. It can effectively improve the effect of multi-label classification. The experimental results show that the scheme proposed in this paper can improve the effect of multi-label classification and is not only suitable for legal documents but also applicable for other text data sets with imbalanced classes and label symbiosis problems.

Classical machine learning works ineffectively when the data set is small. Recently, few-shot learning is proposed to solve this problem. Few-shot learning models a few samples through the prior knowledge. We could divide few-shot learning into various categories depending on where the prior knowledge is extracted from. There are mainly three classes in this paper: (i) the prior knowledge extracted from the labeled data; (ii) the prior knowledge extracted from a weakly labeled or unlabeled data set; (iii) the prior knowledge extracted from similar data sets. For the convenience of searching corresponding few-shot learning methods in a certain domain, based on the above classification, we further classify few-shot learning models into ones which are applied to the image domain and the other which are applied to the text domain. With this taxonomy, we review the previous works on few-shot learning and discuss them according to these categories. Finally, present challenges and promising directions, in the aspect of few-shot learning, are also proposed.

Fire is a common disaster which frequently causes serious losses. If it can be found in time in the early stage, the loss can be greatly reduced. Traditional computer vision algorithm segments fire areas via binary processing whose disadvantage lies in its the slow speed and low accuracy, failing to meet the requirement of real-time detection of fire areas. With the development of neural networks, deep learning has been applied in the field of computer vision widely. In this paper, we investigate the recognition algorithm for fire images in YOLOv4 environment based on CSPDarknet-53 framework. Firstly, collected images are pre-processed to get the dataset conforming to the specification image production. Then the server is used to train the model to realize recognition, detection and early warnings of fires. At the same time, the system will be further completed to change the current situation of fire image recognition and real-time detection, and improve working efficiency.

With the rapid development of science and technology in China, the continuous progress of artificial intelligence technology has driven the development of all walks of life. The effective application of artificial intelligence system in the field of education and teaching presents a very significant application advantage, and has a profound impact on the whole field of education and teaching. The deep integration of artificial intelligence technology and education has expanded the function of education, improved teaching efficiency and education management services, and promoted the reform of education and teaching. This paper discusses the application of artificial intelligence technology in education and teaching by introducing the meaning of artificial intelligence, its development status in China, and the relationship between artificial intelligence and education and teaching.

Spectrum sensing is the most important part of cognitive radio technology, and it is a necessary condition for spectrum sharing. As the radio environment is constantly changing, it is critical to study how to improve the learning ability of cognitive users through machine learning. Based on learning the historical spectrum sensing data collected by the cognitive users, outputting the spectrum resources available in the current spectrum environment is of great significance for improving the accuracy of spectrum sensing, the efficiency of spectrum sharing, and reducing the impact on authorized users. In this paper, a spectrum sensing model based on Autoregressive Integrated Moving Average model (ARIMA) and Long Short-term Memory (LSTM) is proposed. Moreover, the attention mechanism was added to the LSTM, and the retraining process of the model was optimized through feedback records. Experimental results show that the improved LSTM model has better performance in spectrum prediction than traditional machine learning algorithms.

The visible-infrared person re-identification task remains a challenging issue due to large intra-modality and cross-modality variations. In order to reduce these differences, we propose a novel network model for the visible-infrared task, namely, the tri-modal quadruple constraint network (TQC-Net). First, the TQC-Net converts RGB images to grayscale images through grayscale transformation. The grayscale image loses color information, but is visually more similar to the infrared image. Therefore, it can be used as a bridge between infrared and RGB images. By introducing the grayscale modality, the three-modal image group of the pedestrian is then input to the proposed network, which presents a three-stream framework. In addition, to further reduce the difference between the modalities, we design a hetero-center loss based on a quadruple group and the center loss to train the re-identification model. Extensive experiments are conducted on two datasets, including the SYSU-MM01 and RegDB datasets. The experimental results demonstrate the superiority and effectiveness of the proposed TQC-Net over the state-of-the-art approaches.

With the rapid development of online education, extensive data records from online education are accumulated in large quantities, therefore the educational evaluation industry is of great potential. Cognitive diagnosis based on machine learning has drawn considerable attention from both the research community and industry, and a lot of works have been proposed. However, many models ignored the point that different knowledge concepts have different important degrees on each exercise. In this paper, we propose the Continuous Weighted Neural Cognitive Diagnosis (CWNCD) model, which is extended from the Neural Cognitive Diagnosis (NCD) framework, a cognitive diagnosis framework based on neural network, to get a more accurate diagnosis result and ensure its interpretability. Specifically, we added information about the importance degree of different knowledge concepts in each exercise for modeling their interactions, in which case we can more comprehensively model the cognitive level of a student. Extensive experiments conducted on real-world datasets show that the CWNCD model is feasible and obtain excellent performance. Finally, the possible future research directions are discussed.

Machine learning algorithms based on deep neural networks have achieved remarkable results and have been widely used in different fields. Currently, it is necessary to fully solve privacy issues in machine learning systems, and be able to efficiently encrypt and decrypt data involved in calculations. This article aims to provide the intersection of these two fields, and highlight the technologies used to protect data and models. It introduces the privacy protection capabilities of homomorphic encryption for smart algorithms and efficient data encryption and decryption algorithms, and the benefits of user-oriented privacy protection technologies. The research has certain reference significance.

Clustering is a fundamental task in the computer vision and machine learning community. Although various methods have been proposed, the performance of existing approaches drops dramatically when handling incomplete high-dimensional data (which is common in real world applications). To solve the problem, we propose a novel deep incomplete clustering method, named Deep Incomplete Clustering with Distribution Preserving (DICDP). To avoid insufficient sample utilization in existing methods limited by few fully-observed samples, we propose to measure distribution distance with the optimal transport for reconstruction evaluation instead of traditional pixel-wise loss function. Moreover, the clustering loss of the latent feature is introduced to regularize the embedding with more discrimination capability. As a consequence, the network becomes more robust against missing features and the unified framework which combines clustering and sample imputation enables the two procedures to negotiate to better serve for each other. Extensive experiments demonstrate that the proposed network achieves superior and stable clustering performance improvement against existing state-of-the-art incomplete clustering methods over different missing ratios.

Federated learning is a distributed machine learning method to solve the problems of ‘data islands’ and privacy protection. Now it has become one of the research hotspots in the field of learning. Like many learning methods, Federated Learning is a data-driven learning framework. When facing the challenges of devices heterogeneity, non identically and independently distributed (Non-IID) data and data’s security, it could not be processed simply and abstractly like other learning paradigms. This paper briefly introduces the definition of Federated Learning and the challenges faced by all parties, and mainly summarizes the client scheduling strategies in Federated Learning in recent years. Client scheduling is an important part of the aggregation strategy of Federated Learning. At present, it is very difficult to reduce resource consumption and make the joint model more excellent and personalized. Client scheduling strategy needs to balance the relationship between optimization objectives and task objectives. The summary will help us understand the development situation of the current field and provide a clear direction for future research.

Kinematic parameter error is the main factor that affects the accuracy of industrial robots. The accuracy of robots can be effectively improved through calibration technology. The error model based method is one of main calibration methods for calibrating robots. This paper presents the error sensitivity based redundancy analysis for kinematic calibration of industrial robot. The pose error model is firstly established based on the M-DH model. By introducing the concept of error sensitivity, the effects of kinematic parameter error on the end pose error of industrial robot is analyzed. The redundant parameters of pose error model are analyzed based on the error sensitivity. Secondly, the effectiveness of the non-redundant pose error model is verified through experiments. The average comprehensive position error of Staubli TX60 robot is reduced by 88.7%. The average comprehensive attitude error of Staubli TX60 robot is reduced by 75.4%.

Tibetan linguistics has a long history and has formed a relatively complete traditional grammar system, but there is no recognized and relatively complete modern formal grammar framework of Tibetan. In order to inherit, expand, extend and deepen the traditional Tibetan grammar and its formal requirements, this paper makes comparative study on Fillmore case grammar with traditional Tibetan grammar, and introduces a new Tibetan grammatical unit called case structure, and demonstrates that the case structure plays an important role in Tibetan syntactic and semantic analysis. Different case structures of the same verb form Tibetan sentences, and one Tibetan sentence can be uniquely decomposed into one or more case structures. Case structure not only carries the syntactic structure of the sentence, but also reflects the semantic components of the sentence. Therefore, under this grammatical framework, Tibetan syntactic and semantic analysis can be studied in an integrated way. This opinion has some certain theoretical significance for Tibetan grammar research and Tibetan natural language processing.

Sudden precipitations, especially heavy ones, usually bring troubles or even huge harm to people’s daily lives. Hence a timely and accurate precipitation nowcasting is expected to be an indispensable part of our modern life. Given that the current precipitation nowcasting methods are based on radar echo maps, the ZR relation that transforms radar echoes into precipitation amounts is crucial. However, traditionally the ZR relation was typically estimated by location-dependent experiential formula which is not satisfactory in both accuracy and universality. Therefore, in this paper, we propose a deep learning based method to model the ZR relation. To evaluate, we conducted our experiment with the Shenzhen precipitation data as the dataset. We introduced and compared several deep learning models, such as CNN, LSTM, and Transformer. The experimental results show that Transformer + CNN has a higher prediction accuracy. Furthermore, to deal with the unbalanced datasets and emphasize on heavy precipitation, we tried to use the SMOTE algorithm to expand heavy precipitation samples, and it shows that it can effectively improve the prediction accuracy of heavy precipitation. Similarly, we also tried to use a customized loss function to enhance the weight of heavy precipitation samples during model training, and it also demonstrate that it can achieve a better accuracy of heavy precipitations. Both approaches can improve the prediction of heavy Precipitation samples by more than 30% on average.

It is of great significance to make full use of the complementary advantages of different modality imaging information for improving the accuracy of tumor segmentation and formulating precise radiotherapy plans. This paper proposed a multi-tasking parallel training method, which combined the attention mechanism of specific tasks to mine the effective information of different modals. It has three parallel learning networks based on parameter sharing, including CT segmentation network, MRI segmentation network, and the joint learning network of similarity measurement between CT and MRI images. CT and MRI segmentation networks learned their specific task features, and used the attention module of specific tasks to enhance the utilization of effective features while learning shared features. The similarity measurement learning network jointly learned the similarity between CT and MRI images, and combined the specific task features shared by CT and MRI segmentation networks to segment multimodal tumor images. Comparing the results of single-modal and multi-modal tumor image segmentation, it is proved that multi-modal segmentation can provide more abundant features and effectively locate the tumor location, especially in the fuzzy adhesion region of the tumor boundary. In addition, other multi-modal image segmentation methods were compared, and the results also prove that the multi-task learning method is suitable for multi-modal image segmentation and has achieved better segmentation results.

In order to meet the demand for passenger transport and improve the overall operation efficiency of the integrated passenger transport system in the region within the urban agglomeration, the optimization method of the layout of integrated passenger transport terminals in the urban agglomeration are researched in this paper, and the Beijing-Tianjin-Hebei urban agglomeration is considered as our research object. Firstly, the influential factors of the location and layout of the integrated passenger transport terminal are qualitatively analyzed. Secondly the degree of charm indexes are defined, and next the selection model of alternative points based on the degree of charm of the terminal is built, and then the degree of charm value is calculated comprehensively by the analytic hierarchy process (AHP) and multi-level fuzzy evaluation method. Then, on the basis of determined the alternative points of the terminal, the layout of the terminals is optimized based on the P-median location model, and Microcity software is used to solve the model. Finally, an instance is given to prove that the model method has a certain guiding effect on the layout optimization of urban agglomeration comprehensive passenger terminals.

With the development of Web2.0, huge amount of text information is produced. It’s important to extract useful information from data. This paper systematically analyzes the main research progress and methods of named entity recognition (NER), and grasps the development context to help researchers quickly understand NER. [Method/process] We select representative literature for review, summarize and comb the mainstream methods by bibliometrics and literature research, and count the keywords of relevant papers in Web of Science to support this view, and finally summarize the applications and the development trends of NER. [Result/conclusion] Research shows that common recognition methods include rule-based, statistics-based, hybrid methods, and more and more tend to integrate multiple methods; in recent 5 years, hybrid and joint models based on deep learning are currently dominating the latest technology.

Mobile Edge Computing (MEC) marginalizes the computing resources of the core network. So far, most of results have only focused on small number of users or servers. In this paper, a large-scale MEC wireless network is considered with multi-user multi-MEC servers. The random task calculation with normal distribution is proposed. The joint task offloading and resource allocation is investigated by maximizing the average user offloading utility. The delay and energy consumption offloading benefits is formalized as a mixed integer nonlinear program (MINLP) which is general difficult. Our method is to decompose the original problem into resource allocations (RA) under the fixed offloading decision, and the offloading decision (OD) under the optimal resource allocation. RA is further consisting of one convex optimization and quasi-convex optimization. OD is then solved by using a heuristic search algorithm based on simulated annealing principle. The total simulation time complexity is polynomial in the numbers of users and servers. Simulations show that the proposed algorithms performs closely to the optimal solution and achieve higher system utility than traditional methods in small-size networks. Interestingly, it is also applicable for large-scale networks consisting of more than one thousand users. This provides an efficient way for MEC in practical applications.

In scientific research, it is very important to quickly and accurately retrieve relevant important documents. This paper sorts the keywords according to their importance and calculates the fuzzy membership degree of keywords; The literatures with the same keywords enter the recommendation set, and two kinds of experiments are carried out on the literatures entering the recommendation set: one is to use vague similarity technology to recommend documents according to keywords; Second, on the basis of the first experiment, vague similarity technology is used to fuse the cited numbers and downloads for literature recommendation. Experiments show that the recommendation algorithm proposed in this paper is reasonable and effective, and can give a reasonable ranking, which meets the needs of researchers to retrieve literatures.

Under the trend of the new national energy security strategy, my country’s comprehensive energy development and market competition have become increasingly fierce. This has also led grid companies to continue to innovate comprehensive energy coordination, utilization, and service models. In order to meet the needs of the times, this article focuses on comprehensive The background of energy research puts forward a plan for a provincial integrated energy service platform based on SCADA. The platform is based on CPS and takes the construction of smart energy interactive applications and business management applications as the main line to connect with power customers, energy service providers, government departments and other parties. The platform provides comprehensive energy services externally, and internally relies on professional customer services and diversified service content to promote extensive interconnection and in-depth perception of various energy facilities on the customer side with the power grid. Based on source-network-load-storage collaborative services, it aggregates user resources Implement demand response to support the sustainable development of traditional power supply services and the high-quality development of emerging integrated energy businesses.

At the time of the rapid development of the national economy, various types of emergencies occur frequently, and it is urgent to design a reasonable and effective control model to evaluate and control the scope and extent of the incident. Based on a comprehensive analysis of the characteristics of emergencies, this paper proposes a process oriented management and control model consisting of three sub-modules of data support, intelligence control and event management, and proposes a guarantee strategy for the realization of key aspects of the model. The control activities of the event provide a reference for practical work.

The analysis of factors influencing the carbon emissions of civil aviation transport is of great significance to the low-carbon development of civil aviation transport. In order to quantify the relationship between the change of carbon emission of civil aviation transportation and each influencing factor, this paper decomposes the total carbon emission of civil aviation transportation into four influencing factors through LMDI model. Among them, population size, GDP per capita and turnover per unit of GDP contribute to the growth of carbon emissions, with contribution rates of 5.04%, 81.73% and 21.03%. Respectively, while energy consumption per unit of turnover inhibits the growth of carbon emissions, with a contribution rate of −7.81%. In order to deeply study the change response relationship among the influencing factors of civil aviation transportation carbon emission, the regression equation between civil aviation transportation carbon emission and each influencing factor was obtained by constructing STIRPAT model and ridge regression analysis. Finally, based on the analysis of each influencing factor, targeted suggestions and measures are proposed for energy saving and emission reduction in civil aviation transportation industry.

Campus network users have an increasing demand for file storage and sharing, and the traditional storage and sharing methods cannot adapt to this demand. This study proposes a campus network cloud disk system based on cloud storage, which uses a network asynchronous communication mode to cope with the high load of user concurrency. Besides, the file hashing algorithm is used to solve the problem of multiple copies of files stored on the network, which enables multiple users to share one copy. Symmetric key algorithm is used for file sharing and inexpensive distributed storage is used to facilitate storage space expansion. Finally, Hadoop technology is used for implementation. This campus network cloud disk system adapts to the characteristics of high network bandwidth and low egress bandwidth of campus network, and solves the problem of remote file storage and sharing for teachers and students.

The content of dissolved oxygen in water is an important index to detect and evaluate water quality, and the change of its concentration is greatly affected by algae factors, and water temperature is an important factor to affect algae reproduction. The study of the relationship between water temperature and dissolved oxygen can play a better guiding role in water quality evaluation. In this paper, the web crawler was designed to crawl the water quality monitoring data of the relevant monitoring waters from the Internet, and then the correlation analysis was conducted after the pretreatment. Correlation analysis is mainly divided into three steps: Firstly, analyze the correlation between the obtained water temperature data and dissolved oxygen data in the time domain; Secondly, from the statistical point of view of correlation analysis, perform a regression analysis and get the correlation coefficient; Finally, use the correlation analysis results to predict the future data. The experiment shows that the established correlation analysis model has a good effect on the prediction of dissolved oxygen concentration.

With the improvement of social informatization and the popularization of Internet of Things devices, the scale, complexity and diversity of data are currently growing rapidly, and traditional storage solutions have been unable to meet the complex and diverse applications and large-scale new storage requirements. Existing storage solutions still have deficiencies in data compression and adapting to the diversity of system architectures, resulting in a large waste of storage space resources, which in turn increases the total cost of ownership of platform data. Therefore, this paper will study the data compression strategy of database file storage, and propose a high-relevance mode access data compression method. The data request of the write-only instance of the database hosted on the cloud platform is aggregated with the system workload. The data stored in the write-only instance is compressed, which improves data storage efficiency and storage space utilization. The method was validated using data in real enterprise scenarios. The experimental results show that the proposed method has a certain degree of improvement in storage space utilization compared with the original method.

Against the background of the era of big data and cloud computing, cloud platforms have become the first choice for data storage. While cloud platforms also face serious threats to data security and user privacy. At the same time, users will be faced with the problem of how to efficiently retrieve ciphertext on the cloud. Searchable encryption allows users to directly retrieve ciphertext data on the cloud through keywords, which provides an effective method for solving efficient retrieval of ciphertext on the cloud. As an important branch of searchable encryption technology, proxy re-encryption can realize the sharing of ciphertext among multiple users. However, most of the existing schemes have security shortcomings or functional defects, such as inability to resist keyword guessing attack, and do not support search authority authorization. A new designated-server proxy re-encryption with keyword search scheme was proposed, which not only can resist keyword guessing attack, but also realizes the function of sharing search authority. Finally, we give the performance and security analysis. The experimental results show that our scheme is superior compared with other related schemes.

The consortium blockchain is deployed in many key industries because decentralization is irreversible and traceable. However, the storage system of the alliance chain usually adopts a full copy method, which leads to an explosive increase in storage overhead over time. What is worse, this will become a hidden threshold for companies to join the alliance to a greater extent and then evolve into an industry-centric model again. In this paper, we propose a novel storage solution named HotLT, which adopts a distributed storage system coding scheme to reduce the storage overhead of the alliance chain and enhance scalability. Secondly, to solve the bottleneck of concurrent access to encoded data and improve the average decoding speed, HotLT dynamically adjusts the generating matrix during the encoding process to adopt different encoding strategies for data blocks of different hot. Finally, the data access frequency is further divided, low-complexity coding is performed on data with high access frequency, and high-complexity coding is performed on cold data. Compared with the encoding time of traditional LT encoding, the experimental results show that the encoding time of HotLT is almost the same, and its average access speed gradually approaches the speed of direct reading as the proportion of access hot data increases.

With the advent of the era of big data, data has penetrated into every aspect of social life and become an important production factor in various industries. However, while providing convenience to our life, massive information also brings difficulties to information retrieval, which has problems of low retrieval efficiency and poor retrieval accuracy. Therefore, how to improve the efficiency and accuracy of information retrieval has become a key technical problem to be solved. In this paper, we propose a key word information retrieval scheme based on inverted index, which uses the inverted index segmentation algorithm based on keywords to realize the block retrieval of information, improves the efficiency of information retrieval, and improves the accuracy of retrieval content through reasonable error correction of keywords. The results of evaluation indicate that compared with the mainstream information retrieval methods, the keyword information retrieval scheme based on inverted index can carry out faster and more accurate information retrieval. Experimental results show that the keyword information retrieval scheme with inverted index can further improve the efficiency and accuracy of information retrieval.

Aiming at the characteristics of high-speed movement of SIN network nodes, this article studies the invulnerability of SIN network from the perspective of topological structure. According to the periodicity of satellite constellation, a satellite cycle is divided into multiple time slices and optimized respectively. Taking network redundancy as the optimization goal, the main consideration is the full connectivity of network nodes, the number of nodes and links, and the node load, etc., to construct a network topology invulnerability optimization model. Model solving is an NP-hard problem, this paper proposes a neighbor immune algorithm (NIA) based on simulated annealing update. This method applies improved simulated annealing algorithm (ISAA) before neighbor immune algorithm falls into the local optimum, replacing part of the antibody, so that the population continues to evolve to a better solution. This strategy overcomes the shortcomings of neighbor immune algorithm that it is easy to fall into the local optimum at the later stage of the iteration, and at the same time improves the convergence speed. Finally, the simulation is based on the Iridium constellation with 66 low orbit (LEO), experiments show that the improved algorithm effectively optimizes the optimization effect of the original algorithm, and can obtain a topological structure with good invulnerability.

With the implementation of the “Made in China 2025” plan and the promulgation of the “Renewable Energy Law”, China’s new energy equipment manufacturing industry has developed rapidly. At the same time, it has also revealed that China’s new energy equipment manufacturing enterprises have issues such as a high degree of external dependence on key technologies and low-level international competitiveness. Facing the complex international market environment, Chinese new energy equipment manufacturing enterprises urgently need to improve their independent innovation capabilities. This paper establishes the driving factors of independent innovation by sorting out domestic and international independent innovation-related research and combining the opinions of new energy equipment manufacturing experts. By using the exploratory factor analysis, five dimensions that affect independent innovation capability are extracted: independent innovation input capability, external environmental support, internal environmental support, knowledge management capability, and independent innovation output capability. Finally, it establishes the structural equation model of the driving factors of the independent innovation capability of new energy equipment manufacturing enterprises and proposes the driving path of the independent innovation capability of new energy equipment manufacturing enterprises. The research results show that both independent innovation input ability and knowledge management ability can directly have a positive impact on independent innovation output ability, and the influence of knowledge management ability is more significant; while the external environmental support and internal environmental support influence the enterprise’s knowledge management ability Indirectly affect the output capacity of independent innovation.

The current research on cross-platform instant messaging user association is mainly divided into two categories: based on user attributes and based on user behavior. Methods based on user attributes mainly identify users through multiple attributes such as user name and multi-platform user information association based on cell phone number, but user association is not possible when multi-platform user information is inconsistent and users do not grant their address book permissions. Methods based on user behavior mainly calculate the similarity between user trajectories features such as geographic location frequency and co-occurrence, but this method lacks the user’s information, which leads to the inability to fully excavate the sequential features of the trajectory and affects the accuracy of trajectory matching. In order to solve this problem, this paper proposes a cross-platform instant messaging user association algorithm based on temporal trajectories (CPTrajst). We firstly place probes in the area where the target may appear so as to obtain user information, gets user trajectory, then processes the trajectory and performs two trajectory matches, finally associate users of different platforms whose trajectories match, thus increasing the accuracy and reliability of cross-platform instant messaging user association. We conducts specific experiments for users of WeChat and Momo, the most commonly used instant messengers in China. The results show that we can achieve reliable association for these two types of instant messaging users and the user association accuracy can reach 99.5%, which is better than the existing user association algorithms based on trajectory matching.

In high-density residential areas, intelligent express delivery is often used to solve a large number of express parcels that need to be delivered in a short time. It effectively improves the efficiency of parcel delivery by means of information technology. However, with the continuous development of e-commerce, the logistics demand of the last kilometer delivery is constantly changing, and the problems of intelligent parcel cabinets are also exposed. For example, the express cabinet for parcel delivery is far away from the consignee, and when there are many parcels to be picked up, the consignee takes multiple express routes circuitously, and the inability to pick up all express shipments at one time causes the consignee to pick up the parcels in time and occupy the express cabinet for a long time, which affects the delivery of parcels. The article is clustered based on two main factors: the volume of express delivery and the walking distance to the delivery point. Take the area with a large logistics delivery volume as the center point. Based on the adaptive K-means algorithm, the best combination of package sorting is constructed so that multiple packages of the same address are stored in the same area of the smart express cabinet as much as possible. After determining the best possible delivery location of express parcels, a dynamic optimization model of the parcel delivery location is constructed based on the deep neural network (DNN) algorithm. When the logistics demand fluctuates sharply, it can still effectively allocate massive express parcels to the optimal delivery Click in. In this paper, an empirical study is conducted with the Beijing Mining Community as an example. The results show that the improved DNN algorithm can effectively improve the efficiency of parcel delivery based on smart express cabinets.

Practical warehouse system design can effectively improve warehouse management efficiency. Based on the characteristics of the modern warehouse system, it designs an intelligent warehouse system using Radio Frequency Identification (RFID) tag technology and an intelligent recommendation algorithm. The system includes an RFID-based warehouse management module, an intelligent recommendation module comprising a hybrid recommendation algorithm, and a database module. The hybrid intelligent recommendation algorithm consists of K-means clustering analysis, cosine similarity algorithm, and Markov chain algorithm. Firstly, K-means determines the cluster centroid and then recommends the shelf strategy according to the cosine similarity between the goods. After reaching a specific base, the intelligent recommendation algorithm based on Markov chain will analyze the warehousing habits and rules to deal with the instability caused by the frequent change of mass center after the increasing number of samples. Further, the system allows multi-platform operation of web client and WeChat Mini Programs and it shows the superiority of the system.

Land cover change is a hot topic in the interdisciplinary research of global change and land science. The existing spatial visualization methods based on remote sensing images have the advantages of wide detection range, strong timeliness and objective reflection of land surface changes. However, the data display mode is single and the interaction is weak, the reading threshold is high, and the visual analysis of land use statistics data is insufficient. This paper collects and collates land change data and social and economic data from 2009 to 2016 in China. Firstly, Echarts and other tools are used to achieve visual representation of data. Then the impact of social and economic development needs on land resource utilization is studied. Finally, a prediction model of land use data change is established. In conclusion, this paper presents an effective visual data analysis method according to the characteristics of land use data, which can assist land managers to understand and analyze data and provide scientific basis for their decision-making activities of land use.

With respect to the training program of software engineering specialty, this paper puts forward the comprehensive training platform of Chinese character dictation competition based on Android, and clarifies the purpose and main contents of the comprehensive training platform. Based on the hardware and software facilities of the campus, the C / S model training platform, the use of Android integrated training platform to achieve the Chinese character writing, clearing, timing and other functions, managers of the entire process of the game management, including the participating teams and players, the administrator can simultaneously obtain the client input Chinese characters, and display on the big screen, the judges then give the results after the score and statistical display. The whole integrated training platform is light and practical.

Link prediction aim is to use known information of network to infer missing edges, identify spurious interactions, evaluate network evolving mechanisms, and so on. Currently, with the development of deep learning technology, many neural network-based link prediction algorithms have emerged. However, these existing algorithms, due to the introduction of many parameters, are too computationally expensive to efficiently process large-scale network data. In fact, based on the belief that network nodes with a great number of common neighbors are more likely to be connected, many similarity indices have achieved considerable accuracy and efficiency. Moreover, the method based on structural similarity, simple and applicable, low computing cost, and high prediction accuracy, can quickly process large-scale data. Inspired by the idea, in this paper through the analysis of the network structure index shows that the clustering and assortative coefficients can reflect the similarity between nodes. To this end, this paper integrates the two to form a new link prediction algorithm based on the overall characteristics of the network structure, which realizes the fast and efficient prediction of the missing links in the network. Experiments on 10 real-world networks show that the method is highly accurate and robust compared with baseline.

Taking the Batong Line of Beijing subway as our research object, it includes nine stations, such as Communication Univ. of China Station, Shuang Qiao Station and soon on. The attraction range of passenger flow of the subway station is a circle which takes 800 m as the radius. The surrounding land use types of the subway station and its corresponding area within its attraction limit are determined based on the remote sensing images of each site. The prediction model of the passenger flow of each subway station and the area of different land use types surrounding it is established by using regression analysis. It is found that the average daily passenger flow on weekdays and the average daily passenger flow on weekends of each subway station are positively correlated with its surrounding total land use area. The passenger flow in and out of the subway station at different time periods has different relationship with the surrounding land use types and their corresponding area. Therefore, the average daily passenger flow and passenger flow in different time periods of the subway station can be predicted by its surrounding different land use types.

With the rapid development of Internet of Things (IoT) and 5G, mobile edge computing is gaining popularity for its low computation latency, bandwidth costs and energy consumption. In mobile edge computing, the placement of edge servers is one of the most significant problems and attracts worldwide attention. However, two major problems of edge server placement: high access delay and unbalanced workload of edge servers, have not been completely solved yet. To better solve these two problems, this paper proposes a new Spectral-Clustering-based Edge Server Placement (SCESP) algorithm, which can effectively reduce the access delay and make the workload of each edge server more balanced. In the evaluation, we use the Shanghai Telecom’s base station dataset to test the performance of SCESP and extensive experiments demonstrate the superior performance of SCESP in reducing the access delay and balancing the edge server workload.

Cloud storage service has significant advantages on both cost reduction and convenient data sharing. It frees data owners from technical management. However, it poses new challenges on privacy and security protection. To protect data confidentiality and privacy of users against malicious entities in the cloud, fine-grained data access control in cloud storage has become a challenging issue and draws considerable investigation. Ciphertext-policy attribute-based encryption (CP-ABE) is a promising cryptographic technique to address the above issue. In many scenarios, access policies are associated with privacy and sensitive information of users which needs to be preserved from disclosure. However, existing schemes cannot simultaneously support time-sensitive data publishing and attribute information preservation. In this paper, we propose a privacy-preserving time and attribute factors combined cloud data access control with computation outsourcing scheme (named PTAC). To preserve attribute privacy, we design a dual access policy tree mechanism where one access policy tree is public and another is sensitive and hidden. Moreover, time-sensitive data publishing can be achieved by combining CP-ABE with timed-release encryption. By using edge computing and cloud computing, we also outsource partitive computational cost of encryption and decryption to third parties. Extensive security and performance analysis demonstrate the security and efficiency of our proposed scheme in cloud storage. As a result, valuable attribute information in the access policy can be preserved in case of disclosing to unauthorized recipients.

With the rapid development and wide application of 5G networks, artificial intelligence, and big data, traditional cloud computing cannot handle the massive data generated by edge terminals. Edge computing has gradually been widely used as a computing method close to objects. However, due to the open features of edge computing, such as content perception, real-time computing, and parallel processing, the data security issues that already exist in the cloud computing environment have become more prominent. Data security protection capability evaluation is an important part of the improvement of data security capabilities. However, current protection capability evaluations are mostly qualitative evaluations and lack quantitative evaluation models. Aiming at the edge computing network architecture, this paper proposes a data security protection capability evaluation model based on weight presets. By studying the edge architecture data security protection capability score and cost curve, the data security protection model is adaptively selected. Experiments show that the method can quantitatively calculate the data security protection capability in the edge computing environment, and can guide the construction of the protection model through the score-cost curve.

With the rapid development of the network, various businesses continue to appear, and the number of cyberspace protection targets is increasing exponentially. How to find out the family background, recognize the risks, find out the loopholes and perceive the network security situation is an urgent problem to be solved. This paper briefly introduces the related concepts of cyberspace and cyberspace surveying and mapping, then analyzes the technical system of cyberspace surveying and mapping, and puts forward the iterative evolution technical framework of five links: target classification, collaborative detection, fusion analysis, visual mapping and system application. Among them, target classification is the basis of cyberspace surveying and mapping, and collaborative detection and fusion analysis are the key of Surveying and mapping, Visual mapping is the efficiency presentation of Surveying and mapping, and system application is the final surveying and mapping goal. Several stages of cyclic evolution make the surveying and mapping ability of cyberspace rise spirally. The purpose of this paper is to quickly and accurately find all kinds of asset targets in the network, timely perceive asset risk, and lay the foundation for the research progress of cyberspace mapping.

The Industrial Internet of Things (IIoT) enables interconnection and intelligent collaboration among basic industrial production factors which include human, machine, thing, method and environment. In the current IIoT applications, it is difficult for collaborative optimization and unified management and control of industrial production factors. Applications and industrial production factors are tightly coupled, so that many industrial software applications generally have the problems such as high degree of customization and difficult replication and promotion. In this paper, focusing on the most basic industrial production factors, we propose the solution and system architecture of software-defined Industrial Internet of Things (SD-IIoT) based on the ideas and technologies of software definition and Cyber-Physical System (CPS). The principle of SD-IIoT is introduced from the perspective of cyber-physical space mapping. On basis of the digital twin models of industrial production factors, the system architecture of SD-IIoT is designed, which decouples upper-level industrial applications from the underlying industrial production factors. Furthermore, the software definition mechanism based on industrial information model is proposed to abstract and describe industrial production factors with semantic technology to implement the virtualized modeling. The SD-IIoT paradigm can maximize the utilization of resources, and achieve the modular management, on-demand reusing, dynamic reconfigurability and efficient collaboration of industrial production factors, thus improving overall service capability of IIoT.

Particle Swarm Optimization (PSO) algorithm is widely used in cloud computing task scheduling. As PSO algorithm tends to fall into local optimum and has poor convergence accuracy in the later stage, as a result, in this paper i integrates Simulated Annealing algorithm (SA) into PSO algorithm and adopts a strategy of mixing random and nonlinear decreasing inertia weight. At the same time, the idea of “accept the bad solution with a certain probability” in simulated annealing is used to improve the global search ability of the algorithm. Finally, the chaotic disturbance strategy is added to make the algorithm search for a better solution as far as possible, so as to improve the convergence accuracy of the algorithm in the later stage. The improved PSOSA algorithm and the standard particle swarm optimization algorithm are applied to the task scheduling test in the cloud environment through the Cloudsim simulation platform. The results show that the improved algorithm can achieve better scheduling results and faster convergence speed.

In order to meet the needs of group enterprises for data privacy protection processing in the internal private network environment, this paper proposes an optimization design method of Privacy Protection System (PPS) based on cloud native. The main goal of this method is to transform structured data using data anonymization techniques to mitigate attacks that could lead to privacy breaches. From the perspective of logical architecture, this method supports the anonymization of data using the privacy protection model and related parameters selected by the user. First, the identifiers are removed from the dataset to be processed, and constraints are imposed on the quasi-identifiers. Further, the algorithms for protecting sensitive properties that require certain assumptions about the attacker’s goals and background knowledge are also supported. In particular, in the process of anonymization, the scheme introduces the methods of utility analysis and risk analysis, so that the anonymization results can be accurately evaluated. Finally, the proposed method allows users to iteratively update the privacy-preserving model and related parameters according to the results of the anonymization evaluation. From the perspective of technical architecture, the proposed method uses Spring Boot as the back-end framework and MyBatis as the persistence layer framework. At the same time, in order to ensure system security requirements, Json Web Token is also used for user authentication. Finally, when designing the system deployment scheme, cloud native technology is introduced to encapsulate system functions into microservice containers, and cluster management tools are used to dynamically manage microservice containers to ensure high availability of the system.

Network traffic, which records users’ behaviors, is valuable data resources for diagnosing the health of the network. Mining anomaly in network is essential for network defense. Although traditional machine learning approaches have good performance, their dependence on huge training data set with expensive labels make them impractical. Furthermore, after complex hyperparameters tuning, the detection model may not work. Facing these challenges, in this paper, we propose Autoperman through supervised learning. In Autoperman, machine learning algorithms with fixed hyperparameters as feature extractors are integrated, which utilize a small amount of training data to be initialized. Then Random Forest is selected as the anomaly classifier and achieves automatic parameters tuning via well studied online optimization theory. We compare the performance of Autoperman against traditional anomaly detection algorithms using public traffic datasets. The results demonstrate that Autoperman can perform about 6.9%, 34.2%, 4.3%, 2.2%, 37.6 % better than L-SVM, NL-SVM, LR, MLP, K-means, respectively.

Recent advancements in human-computer interaction research of Internet-of-Emotion have led to the possibility of emotional communication via the human-computer interface for a user with neuropsychiatric disorders or disabilities. There are several ways of recording psychophysiology data from humans, and in this paper, we focus on emotion detection using electroencephalogram (EEG). Various emotion extraction techniques can be used on the recorded EEG data to classify emotional states. Band energy (E), frequency band energy ratio (REE), the logarithm of the frequency band energy ratio (LREE), and differential entropy (DE) of band energy ratio is some emotion features that previously have been used to classify EEG data in various emotional states. Four different emotion features were analyzed in this paper, classifying EEG data associated with specific emotional states. The results showed that DE is the best choice in Wavelet Transform-Support Vector Machine (WT-SVM) model during the period of training an SVM classifier to be accurate over the whole data sets (16 subjects), and the whole accuracy up to 86.5%, while DE’s classification results are between 73.81% to 97.62%. This phenomenon shows that it is difficult to find features that are generally working well over each subject, and there is also the possibility that the pictures of the International Affective Picture System (IAPS) did not induce strong enough emotions on some subjects making it difficult to classify some emotional states. Based on the result, we only conclude that DE is the optimal choice for the WT-SVM model, and the individual factors will be considered as one of the influencing factors of the emotion classification system in future work.

The emergence of mobile edge computing greatly alleviates the problem of insufficient computing power of mobile devices and does not support high-energy applications. As an important part of edge computing, computing offloading can greatly improve the quality of service through a reasonable computing offload scheme. For time delay sensitive tasks, the time delay of computational offloading under energy consumption constraints is too large, this paper introduces an improved fireworks algorithm based on grouping and classification (GCFA). The problem is modeled as the minimum delay problem under the constraint of energy consumption, and the offloading vector is calculated by GCFA, which transforms the task offloading into the process of fireworks particle optimization. Finally, through experiments, the genetic algorithm (GA) offloading strategy, standard fireworks algorithm (FA) offloading strategy, bat algorithm (BA) offloading strategy and mayfly optimization algorithm (MA) are compared, the average total system cost of GCFA is much lower than that of the other four. The total system cost of GCFA is 20% lower than that of the original. The experimental results show that GCFA has a good performance in reducing MEC time delay and balancing the load of MEC server.

Emerging machine learning methods have become a powerful driving force to revolutionize many industries nowadays, such as banking, healthcare services, retail, manufacturing, transportation. Meanwhile, privacy has emerged as a big concern in this machine learning-based artificial intelligence era. Functional encryption is a new type of encryption primitive in which a secret functional-key allows one to compute a specific function of plaintext from the ciphertext, making it very suitable for privacy protection machine learning scenarios. In this paper, we apply the concepts of decentralized multi-client function encryption to explore a new solution to the privacy-preserving convolutional neural network. The results of the experiment show that our scheme is feasible, and the accuracy of the final model on the test set is 92.1%, which is close to 93.2% of the convolution network connected to plain text.

Space information networks (SINs) has the advantages of low power consumption and high data transmission rate in the process of network information transmission. It is a good solution to realize the systematic application of space information. The establishment of SINs need to generate network topology, and optimize and improve the invulnerability of SINs on this basis. In order to optimize the distributed minimum spanning tree algorithm (DMST) algorithm, an improved approximation algorithm, degree-guarantee minimum spanning tree algorithm (DGMST) algorithm, is proposed. DGMST algorithm adopts advanced data structure, which greatly reduces the time complexity. On the basis of fully considering the limit of satellite node link degree, generating the minimum average edge weight of subgraph and the requirements of topology invulnerability, DGMST algorithm further considers some boundary conditions to ensure the successful generation of the generated subgraph. Finally, it also reduces the average edge weight of the generated subgraph as much as possible, so that the dynamic network structure has better invulnerability and security.

Most of existing language models function by predicting next candidate words based on previous words or contextual statistics. However, they ignore the characteristics of the language itself, such as morpheme including words prefixes, roots, suffixes, etc. which play a significant role in understanding. Tibetan is a language whose complete meaning relies on its functional words such as the suffix of their front adjacent words, where some functions expresses by the radicals, the parts of the Tibetan characters. To utilize the characteristics of Tibetan in a language model, we proposed a novel language model which considers the functional words, especially free-functional words and suffixes of their preceding words. We first construct a Tibetan language model by utilizing the relationship between functional words and the ten explicit Tibetan suffixes, referred to as Tibetan Radical Suffix Unit-Explicitly (TRSU-E). To taking into account of the free-functional words, we improve the TRSU-E model to address the implicit relationship between free-functional words and their front suffix words, which denotes as Tibetan Radical Suffix Unit-ALL (TRSU-ALL). A standard Tibetan corpus is constructed and used for comprehensive analyses and evaluations. Experimental results show that TRSU-E achieves 16.29% and 10.65% relative perplexity reduction compared with two state-of-the-art methods, i.e. RNNLM and Tibetan Radicals Unit (TRU) respectively.

With the technology advancing rapidly, it has become apparent that it is nearly impossible to go without a digital trace when committing a crime. Device such as Xiaomi wristband, keeps the record of a user’s daily activities, heart rate, pressure value, step count, and sleep quality. Due to daily use, these messages are valuable for digital forensic investigators as it may serve as evidence and proofs to a crime, as well as revealing the motives of crime. This paper will demonstrate some useful examples to investigators about the methods to extract and analyze smart watch using Bluetooth tools like Ubertooth One. Experiments about wireless sniffing and wired transmission are given in this paper. We are making a tentative exploration and hoping to provide some useful experience and practice to investigators in the field of forensics.