Advanced Hybrid Information Processing

The existing compressive sensing recovery algorithm has the problems of poor robustness, low peak signal-to-noise ratio (PSNR) and low applicability in images inpainting polluted by impulsive noise. In this paper, we proposed a robust algorithm for image recovery in the background of impulsive noise, called $$\ell _{p}$$ -ADMM algorithm. The proposed algorithm uses $$\ell _{1}$$ -norm substitute $$\ell _{2}$$ -norm residual term of cost function model to gain more image inpainting capability corrupted by impulsive noise and uses generalized non-convex penalty terms to ensure sparsity. The residual term of $$\ell _{1}$$ -norm is less sensitive to outliers in the observations than $$\ell _{1}$$ -norm. And using the non-convex penalty function can solve the offset problem of the $$\ell _{1}$$ -norm (not differential at zero point), so more accurate recovery can be obtained. The augmented Lagrange method is used to transform the constrained objective function model into an unconstrained model. Meanwhile, the alternating direction method can effectively improve the efficiently of $$\ell _p$$ -ADMM algorithm. Through numerical simulation results show that the proposed algorithm has better image inpainting performance in impulse noise environment by comparing with some state-of-the-art robust algorithms. Meanwhile, the proposed algorithm has flexible scalability for large-scale problem, which has better advantages for image progressing.

With the rapid development of network information technology and the popularity of mobile intelligent terminals, Internet industry has experienced rapid growth. As the main force for building socialism, college students have demonstrated unprecedented passion for innovation and entrepreneurship. In the ‘Internet plus’ background, under the innovation and entrepreneurship service platform for college students, resource scheduling is the key to the entire system, affecting the performance of the system. For this reason, this paper proposes a resource scheduling algorithm for college students’ innovation and entrepreneurship service platform based on load balancing, and constructs a dynamic project resource allocation scheme, which improves the utilization of resources. Experimental results verify the effectiveness of the proposed resource scheduling algorithm, the match rate between the project and the university graduates has increased by 30%.

The Internet has become one of the important channels for users to access to information and knowledge. It is crucial that how to acquire key content accurately and effectively in the events from huge amount of network information. This paper proposes an algorithm for automatic generation of network document summaries based on knowledge graph and TextRank algorithm which can solve the problem of information overload and resource trek effectively. We run the system in the field of big data application in packaging engineering. The experimental results show that the proposed method KG-TextRank extracts network document summaries more accurately, and automatically generates more readable and coherent natural language text. Therefore, it can help people access information and knowledge more effectively.

With the rapid development of Internet technology and the national economy, the brand awareness of consumers and businesses is getting stronger and stronger. However, how to identify application solutions for counterfeit and shoddy goods is a serious challenge for commodity brand. This paper designs and implements a two-dimensional code packaging advertising and anti-counterfeiting model based blockchain. It is applied to process the commodity information in the packaging field, based on the decentralization, openness, autonomy, anonymity and non-tamper ability of blockchain, combined with two-dimensional code technology. The model is based on network transmission interaction, has the advantages of high unforgeability, low cost, easy implementation, fast access, etc. It has a good technical reference value for implementing packaging advertising, anti-counterfeiting and blockchain application.

Aiming at the problems such as complex feature extraction, low recognition rate and low robustness in the traditional human action recognition algorithms, an improved 3D convolutional neural network method for human action recognition is proposed. The network only uses grayscale images and the number of image frames as input. At the same time, two layers of nonlinear convolutional layers are added to the problem of less convolution and convolution kernels in the original network, which not only increases the number of convolution kernels in the network. Quantity, and make the network have better abstraction ability, at the same time in order to prevent the network from appearing the phenomenon of overfitting, the dropout technology was added in the network to regularize. Experiments were performed on the UCF101 data set, achieving an accuracy of 96%. Experimental results show that the improved 3D convolutional neural network model has a higher recognition accuracy in human action recognition.

HF cognitive radio is considered to be one direction of fourth generation HF radios. In this paper, we investigate the problem of multi-user HF radio communication by leveraging the techniques of cognitive radio. In the presented system model, we consider the determination of optimal path between two points and propose a channel probing method based on coarse granularity wideband spectrum sensing as well as channel prediction. To cope with the problem of channel selection and link establishment, we adjust the channel selection strategy after every probing based on Stochastic Learning Automata (SLA) learning algorithm. The experimental results show that the channel selection based on SLA learning algorithm is better than random channel selection, and channel selection with predicted wideband spectrum sensing performs better in system performances than no-predicted narrowband spectrum sensing.

Through research on the access control of software defined network (SDN) northbound interfaces, we found that malicious OpenFlow applications (OF applications) abuse the northbound interfaces with ADD permissions, which can cause the controllers function failure and other serious harm or even crash directly. Most previous studies of this issue, such as those resulting in the ControllerDAC scheme, set static thresholds; and did not find effective solutions to those problems. This paper analyzes the characteristics of the input flows and proposes an SDN dynamic access control scheme based on prediction and dynamic adjustment of the load threshold. By examining the access characteristics of the OF application, we use a prediction algorithm to determine whether the application will disrupt the API with ADD permissions. This algorithm enables us to perform targeted dynamic access control for different types of applications. Experimental results show that compared with the aforementioned ControllerDAC scheme, our scheme effectively reduces the malicious flow table rate and limits the delivery of malicious flow tables, and the extra delay generated by our scheme is less than 10%.

To overcome the traditional disadvantages of single source points detection methods in underdetermined blind source separation problem, this paper proposes a novel algorithm to detect single source points for the linear instantaneous mixed model. First, the algorithm utilizes a certain relationship between the time-frequency coefficients and the complex conjugate factors of the observation signal to realize single source points detection. Then, the algorithm finds more time-frequency points that meets the requirements automatically and cluster them by utilizing a clustering algorithm based on the improved potential function. Finally, the estimation of the mixed matrix is achieved by clustering the re-selected single source points. Simulation experiments on linear mixture model demonstrates the efficiency and feasibility for estimating the mixing matrix.

The electromagnetic spectrum is an important national strategic resource. Spectrum sensing data falsification (SSDF) is an attack method that destroys cognitive networks and makes them ineffective. Malicious users capture sensory nodes and tamper with data through cyber attacks, and make the cognitive network biased or even completely reversed. In order to eliminate the negative impact caused by abnormal information in spectrum sensing and ensure the desired effect, this thesis starts with the improvement of the performance of cooperative spectrum sensing, and constructs a robust sensing user evaluation reference system. At the same time, considering the dynamic changes of user attributes, the sensory data is identified online. Finally, the attacker identification and elimination algorithm is improved based on the proposed reference system. In addition, this paper verifies the identification performance of the proposed reference system through simulation. The simulation results show that the proposed reference system still maintain a good defense effect even if the proportion of malicious users in the reference is greater than 50%.

With the popularization of self-driving cars and the rapid development of intelligent transportation, pedestrian detection shows more and more extensive application scenarios in daily life, which have higher and higher application values. It also raises more and more interest from academic community. Pedestrian detection is fundamental in many human-oriented tasks, including trajectory tracking of people, recognition of pedestrian gait, and autopilot recognition of pedestrians to take appropriate response measures. In this context, this paper studies the design and implementation of a pedestrian detection system. The pedestrian detection system of this article is mainly composed of two parts. The first part is a pedestrian detector based on deep learning, and the second part is a graphical interface that interacts with the user. The former part mainly uses the Faster R-RCNN learning model, which can use convolutional neural networks to learn features from the data and extract the features of the image. It can also search the image through RPN network for areas where the target is located and then classify them. In this paper, a complete pedestrian detection system is implemented on the basis of deep learning framework Caffe. Experiments show that the system has high recognition rate and fast recognition speed in real world.

1090ES ADS-B is a new technology in civil aviation, mainly used to monitor dynamic condition of aircrafts and share the flight information. However, the ADS-B signals are often interfered with other overlapped signals during the signal transmission, causing signal overlap and bringing difficulties to signal processing afterwards. This article applies Blind Source Separation (BSS) into the separation process of ADS-B overlapped signals and constructs the ADS-B overlapped signal separation model using Infomax algorithm, in order to separate ADS-B overlapped signals into single-way ADS-B signals and improve signal decoding rate.

When using traditional monitoring technology to monitor the disaster area of potato in Northern Shaanxi, there was a problem of insufficient monitoring accuracy. In view of the above problems, a real-time monitoring technology for potato pests and diseases based on hyperspectral data is put forward. Firstly, the geological environment of the monitoring area is briefly introduced. Hyper Spectral Remote Sensing is used to obtain the hyperspectral data of the damaged area of the potato in the study area, and pretreatment is performed to establish a regression model. Finally, the pre-processed hyperspectral data is obtained. Substituting data into the model, the area of potato pests and diseases in the research area is obtained. The results showed that the accuracy of the method was 20.29% higher than that of the traditional potato pest and disease monitoring technology, and the accurate monitoring of the disaster area was realized. It has practicality and superiority.

For traditional search methods affected by the index build time, resulting in poor search results, a multi-mode retrieval method for big data of economic time series based on machine learning theory is proposed. According to the good extensibility of big data, construct a retrieval model and use binary data conversion methods to match big data. The binary sequence is defined by the relationship between different data, the similarity of data features is calculated, and the candidate candidate sequence is filtered. Data with no similar features are filtered, and each sub-sequence set matching the pattern is given by similarity size. After the threshold is added, on the basis of slightly reducing the filtering amplitude, the calculation of the similarity matching in the big data retrieval process is greatly reduced, and combined with the fixed interval sampling matching method to determine the characteristics of big data, thereby realizing the machine learning theory. The multi-mode retrieval method for big data of economic time series based on machine learning theory retrieval. According to the experimental comparison results, the retrieval efficiency of the method can reach 95%, which provides effective help for large-scale retrieval of massive data.

The parallel prediction model of big data with traditional power load has a low prediction accuracy in different working conditions, so the parallel prediction model of big data for short-term power load is designed. The short-term power load forecasting theory is analyzed, and the short-term power load data are classified to select the short-term power load forecasting theory. The Map/Reduce framework is built on the basis of the theory, and the prediction process is designed through the Map/Reduce framework. The short-term power load data of the subnet and the big data of the short term power load are predicted respectively, and the construction of the parallel prediction model of the short-term power load big data is realized. The experimental results show that the proposed big data parallel prediction model is better than the traditional model, and can be switched under different working conditions, and the deviation between the forecasting curve and the actual load is small, the average deviation is 1.7, and the overall prediction effect is good.

In the context of big data, data fusion in the perception layer of the Internet of Things is extremely necessary. Fusion data can reduce the amount of data traffic in the network, avoid wasting network resources and bring great convenience to users’ observation and analysis. Aiming at the high computational complexity of the data fusion algorithm at the current, an approximate data fusion algorithm for the perception layer of the Internet of Things based on the probability distribution is proposed in this paper. Firstly, the data fusion model of the perception layer of the Internet of Things and the probability distribution model of the node data are analyzed. And then, disturbances are applied to the node data to achieve the purpose of concealing the collected data. Finally, the approximate fusion of data in the sensing layer is achieved by collecting the probability distribution of the data. The experimental results verify the effectiveness of the fusion algorithm and test the influence of the algorithm parameters on the fusion effect, which provides a reference for the engineering implementation of the algorithm.

At present, global positioning system (GPS) is the most widely used positioning technology for outdoors, but when it is indoors, its positioning accuracy will become lower. The anti-jamming ability of radio frequency identification (RFID) is strong, and it can be carried out in bad environment. Because of its advantages of non-sight distance, non-contact, relatively low price and so on, the application of RFID can realize the requirement of high-precision positioning. By analyzing the existing indoor positioning system, an algorithm of selecting the adaptive threshold value in virtual reference elimination (VIRE) is proposed in this paper, it can find the appropriate threshold values for each target to accommodate complex changes in the indoor environment, simulation results manifest its effectiveness.

Conventional direction finding algorithms for wideband signal need to preliminarily estimate the Direction of Arrival (DOA) and power of the noise roughly, and it has large focusing error. In order to solve these problems, a super-resolution direction finding algorithm for two dimensional (2-D) wideband chirp signals is proposed, the Fractional Fourier Transform (FRFT) is applied to focus the energy of the signals in every frequency by the rotational characteristic of FRFT, then algorithm for narrowband signals is used to estimate the DOA, computer simulation results prove the effective of the algorithm.

As the synchronism effect of traditional methods is poor, the research on data synchronism method in heterogeneous database based on Web Service is proposed. Based on the structure of heterogeneous database, the synchronous execution flow is designed, and HTTP based SOAP transport protocols and XML standard are used to encode the data uniformly, the data source can be added or deleted at any moment. By decomposing the SQL statement in synchronous control module of heterogeneous database, the results data returned are converted into XML format with XML conversion function and are returned to uniform Web Service interface. The data synchronization in heterogeneous database based on Web Service is achieved through the local temporary table and its copy of source database. Through the comparative experiment, the following results can be concluded: the synchronism effect of data by data synchronism method in heterogeneous database based on Web Service reached 97%, which met the requirements for dealing with normal business in synchronism application system.

The traditional English course planning information automatic calibration system has poor precision, weak data analysis ability and low calibration accuracy. To solve this problem, the new English course plan information automatic calibration system is designed with big data analysis technology, and the hardware and software parts of the system are designed. Designed with high-precision ARM processor, TA64 embedded tracking chip and CS652 positioning chip, the hardware consists mainly of two types of power supply, calibrator and monitor in series and parallel. The software part is designed with five functional modules: teaching case information collection, information processing, information analysis, information correction and correction structure detection to complete software optimization. The effectiveness of the calibration system has been verified compared to traditional automatic calibration systems. The experimental results show that the system has strong data analysis capability, high precision after calibration, good calibration effect and large development space.

The traditional intelligent drive system of variable speed vehicle has the problem of low precision of driving target, so the intelligent drive system of variable speed vehicle based on big data analysis is designed. The hardware structure of intelligent driving system of variable speed vehicle is designed, and the hardware framework of the system is derived on the basis of the hardware structure, so as to complete the hardware design of intelligent driving system of variable speed vehicle. Respectively for automated driving simulation component library, automotive autopilot system protection module, the automatic speed control driving system design, complete the autopilot system software design, through software and hardware design to realize variable speed motor intelligent automatic driving system design. Experimental contrast can be seen that is based on the analysis of the large data variable speed auto intelligent automatic driving system compared with the traditional automatic driving method, driving the target precision increased by 15%, and has high effectiveness.

The motion feature data capture can well preserve the details of the motion and truly record the trajectory of the motion. It has been widely used in many fields such as virtual reality, three-dimensional games, film and television effects, and so on. With the widespread application of motion feature capture, how to analyze the trend data of sports features has become a hot topic. The main purpose of the trend analysis of the research motion characteristics is to better understand and describe the motion process of the objects so as to manage and reuse the motion capture data in the motion capture database. For the existing motion feature capture data in the motion capture database, the motion feature data behavior is precisely segmented, the motion template is extracted and calculated more quickly and efficiently, the motion behavior is identified, and the motion behavior in the motion sequence segment is automatically identified.

Conventional cloud computing iot encrypted data comparative analysis method to simple encryption scheme comparison, the accuracy of the scheme comparison is not high, for more complex data encryption scheme comparison, comparative stability lower deficiencies, therefore put forward under the cloud computing environment, the Internet of things encrypted data comparison analysis research. Introducing the sliding window technology, build the Internet of things to encrypt data security evaluation mechanism, determine the iot encrypted data comparison analysis algorithm, constructing iot encrypted data comparison analysis model is presented. Run the analysis model, analyze the iot encrypted data comparison scheme, and implement the iot encrypted data comparison scheme analysis. The experimental data show that the proposed scheme is more routine than the conventional scheme analysis, and the stability of the method is maintained at 70%–90%, which is suitable for the comparison scheme analysis of the network encryption data with different difficulty coefficients. The proposed method for comparative analysis of IoT encrypted data is highly effective.

SDN is a new network architecture with centralized control. By analyzing the traffic characteristics of DDoS attack, and using the SDN controller to collect the traffic in the network, the important characteristics such as the IP address entropy ratio and the port entropy ratio related to the attack are extracted. According to the analysis of relevant eigenvalues, the RBFNN algorithm is used to classify the training samples to detect DDoS attacks. Finally, the SDN environment and DDoS attacks are simulated under Ubuntu, and the RBFNN algorithm detection model is deployed in the SDN controller. Compared with BPNN algorithm and Naive Bayes algorithm, it is proved that the algorithm performs DDoS attack detection with high recognition rate in a short time.

Aiming at the problem that the traditional K-nearest neighbor algorithm has a long classification time when predicting Trojan sample categories, this paper proposes a speed-up K-nearest neighbor classification algorithm CBBFKNN for Trojan detection. This method adopts the idea of rectangular partitioning to reduce the dimensionality of the sample data. Combining the simulated annealing algorithm and the Kmeans algorithm, the sample set is compressed and the BBF algorithm is used to quickly classify the sample. The experimental results show that, the CBBFKNN classification algorithm can effectively reduce the classification time while the precision loss is small in IRIS dataset. In terms of Trojan detection, the CBBFKNN classification algorithm can guarantee higher accuracy and lower misjudgment rate and lower missed detection rate in shorter detection time.

To determine the number of signals arriving on an array of sensors correctly is very important for most high resolution DOA (direction of arrival) estimation algorithms, the methods based on information theoretic criteria have good properties when there is a large snapshots, while it always leads to an error in the small snapshots field. A method based on the exact distribution of the eigenvalues of the sampling covariance matrix is proposed in the paper, it makes use of the model of information theoretic criteria at the same time, the new method has excellent performance when the snapshots is small, the computer simulation results prove the effective performance of the method.

The Internet develops rapidly in the era of big data, which can be shown by the widespread uses of image processing software as well as digital images skills. However, there are a large number of redundant images in the network, which not only occupy the network storage but also slow down image search speed. At the same time, the image hash algorithm has received extensive attention due to its advantages of improving the image retrieval efficiency while reducing storage space. Therefore, this paper aims to propose a large-scale image retrieval method based on image redundancy and hash algorithm for large-scale image retrieval system with a large number of redundant images. I look upon the method into two phases: The first phase is eliminating the redundancy of repetitive images. As usual, image features need to be extracted from search results. Next, I use the K-way, Min-Max algorithm to cluster and sort the returned images and filter out the image classes in the end to improve the speed and accuracy of the image retrieval. Fuzzy logic reasoning comes to the last part. It can help to select the centroid image so as to achieve redundancy. The second phase is image matching. In this stage, the supervised kernel hashing is used to supervise the deep features of high-dimensional images and the high-dimensional features are mapped into low-dimensional Hamming space to generate compact hash codes. Finally, accomplish the efficient retrieval of large-scale image data in low-dimensional Hamming of the space. After texting three common dataset, the preliminary results show that the computational time can be reduced by the search image redundancy technology when filter out the invalid images. This greatly improves the efficiency of large-scale image retrieval and its image retrieval performance is better than the current mainstream method.

Eigenvalue-based spectrum detection has become a research hot topic, which can make detection by catching correlation features in space and time domains. However, most existing methods only consider part of eigenvalues rather than all the eigenvalues. Motivated by this, this paper focuses on all the eigenvalues of sample covariance matrix in digital channelized system and proposes an improved sub-band spectrum detection method. Utilizing the distribution characteristics of the maximum eigenvalue of covariance matrix and the correlation of all the average eigenvalues, a better theoretical expression of detection threshold is obtained. The proposed method can not only overcome the affection of noise uncertainty, but also achieve high detection probability under low SNR environment. Simulations are performed to verify the effectiveness of the proposed method.

When the traditional multimedia data security storage method is used to store the electronic commerce data, there are some problems such as insufficient security and slow encryption speed, which often lead to data leakage or loss. A secure storage method for multimedia data based on block chain technology is proposed. This method takes the study of multimedia data in the field of electronic commerce as an example. Firstly, the key technology of this method: block chain and multimedia data are introduced briefly, then the block chain of multimedia data is constructed. Finally, a multimedia data storage model of e-commerce is established on the basis of block chain structure. The model is divided into two parts: multimedia data addition Secret and multimedia data preservation. The protection of multimedia data is realized. The results show that compared with the traditional multimedia data security storage method, the security of this method is increased by 15% and the encryption speed is increased by 2 s. It is proved that this method can effectively protect the media data for many years and reduce the frequency of data leakage or loss.

Traditional fault signal feature extraction algorithms such as autocorrelation analysis algorithm, morphological gradient algorithm and other algorithms have the disadvantage of low accuracy. Therefore, a fault signal feature extraction algorithm based on wavelet frequency shift algorithm and minimum entropy algorithm is designed. Based on the noise removal algorithm of mechanical equipment based on wavelet frequency shift design and the mechanical fault identification algorithm based on minimum entropy algorithm, the two algorithms are integrated to generate the feature extraction algorithm of mechanical rolling bearing fault signal. In this way, the feature of fault signal is extracted, and an example is given. The experimental results of simulation and application environment design show that, compared with the traditional design, Compared with the fault signal feature extraction algorithm, the proposed algorithm can improve the accuracy of the analysis results by about 4% when using the same data.

Aiming at the problem of multi-band motion and multi-joint inflection point of hybrid manipulator, the conventional trajectory prediction algorithm cannot satisfy the fast analysis and accurate control of motion trajectory. This paper proposes a hybrid manipulator running trajectory prediction algorithm based on PLC fuzzy control. Based on newton-andrews law, the dynamic model of hybrid manipulator was built, and the dynamics of hybrid manipulator was analyzed and the dynamic characteristics were determined. PLC fuzzy control unit is introduced, based on the kinematics characteristics of hybrid manipulator, the relevant input and output variables of PLC fuzzy control unit are determined, and the fuzzy strategy is implemented and analyzed. The construction of a hybrid manipulator based on fuzzy control is completed. The test data show that the proposed prediction algorithm is better than the conventional prediction algorithm, and the accuracy is improved by 57.42%, which is applicable to the prediction of the operation trajectory of the hybrid manipulator.

Traditional mechanical parts manufacturing structure design feature model can more complete mechanical parts structure design feature extracting, but for complex mechanical parts for structural design feature extraction, feature extraction error rates higher deficiencies, this proposed complex mechanical parts manufacturing characteristic model building structure design. Based on the ADO.NET structure, the design platform of complex mechanical parts manufacturing structure is built, the constraint equation of feature model is determined, and the framework of complex mechanical parts manufacturing structure design feature model is constructed. The feature model function of design structure design is designed, and the embedding of complex mechanical structure design software is realized by using XTF embedding technology, and the construction of complex mechanical parts manufacturing structure design feature model is completed. The experimental data show that the proposed complex feature model is 14.24% lower than the traditional model, and is suitable for the application of complex mechanical parts manufacturing structure design characteristics.

This paper studies an unmanned aerial vehicle (UAV)-enabled wireless power transfer system (WPTS) for mobile users, in which a UAV-installed energy transmitter (ET) is deployed to broadcast wireless energy for charging mobile users functioned as energy receivers (ERs) on the ground. Different from the most of the existing research on wireless energy transfer, a dual-dynamic scenario is proposed where a flying UAV transmits wireless power to charge multiple ground mobile users simultaneously. To explore the adjustable channel state influenced by the UAV’s mobility, the UAV’s power allocation and trajectory design are jointly optimized. For the sake of the fairness, we consider the maximum of the minimum of the energy harvested among the nodes on the ground during a finite charging period. The formulated problem above is a non-convex optimization on account of the UAV’s power limit and speed constraint. An algorithm is proposed in the paper to jointly optimize power and trajectory. Simulation results indicate our design improves the efficiency and fairness of power transferred to the ground nodes over other benchmark schemes.

Traditional design method of 3 d mechanical parts to complete the design of mechanical parts, but lack of existing design cycle is long, not suitable for mechanical three-dimensional rapid design of the parts and components for manufacturing oriented network collaborative 3 d mechanical parts rapid design technology. Using three-dimensional mechanical parts the construction of the collaborative design platform, and the determination of network collaborative design principle based on the establishment of network collaborative design data transfer mode to complete network collaborative construction of three-dimensional mechanical model of rapid design components; Web-based collaborative design task decomposition, and the conflict of network collaborative design solutions, relying on the online conflict detection, access control, ORG connecting key technology to realize network collaborative 3 d mechanical parts rapid design. Experimental data show that the proposed rapid design technology compared with traditional design, shorten the design cycle by 84.41%, at the same time to ensure accuracy, good design is suitable for mechanical 3 d parts of rapid design.

Data assimilation plays a very important role in numerical weather forecasting, and data assimilation algorithms are the core of data assimilation. The objective function of common data assimilation algorithms currently has a large amount of calculation, which takes more time to solve, thereby causing the time cost of the assimilation process to affect the timeliness of the numerical weather forecast. Aiming at an excellent hybrid data assimilation algorithm-dimension reduction projection four-dimensional variational algorithm that has appeared in recent years, the paper uses the MPI parallel programming model for parallel implementation and optimization of the algorithm, and effectively solves the problem of large computational complexity of the objective function. This effectively not only reduces the solution time of the algorithm’s objective function, but also ensures the effect of assimilation. Experiments show that the speedup of the paralleled and optimized algorithm is about 17, 26, and 32 on 32, 64, and 128 processors, and the average speedup is about 26.

The traditional data extraction method of complex linear pixel image data can not cope with the vibration and noise of data in the process of data vectorization, which causes the problem of low accuracy of the extraction results. To solve this problem, a complex vector image extraction method is proposed. The MATLAB method is used to remove the noise of complex linear pixel images. In this way, the preprocessing of complex linear pixel image data is provided as the condition of segmentation. The two value algorithm is used to segment the complex linear pixel image data, and the minimum value of the target function is calculated. The data curves are drawn according to the calculation results. Vectorization of image data. The experimental results of the simulated application environment design show that the accuracy of the extraction result is about 45% compared with the traditional extraction method when the same image data is used.

Aiming at the insufficiency of the logistics distribution model of traditional agricultural products, this paper puts forward the optimization modeling analysis of the intelligent logistics distribution route of agricultural products under the Internet of Things. According to the logistics distribution model of agricultural products under the Internet of Things environment, the intelligent logistics distribution path of agricultural products is optimized and modeled and analyzed. The objective function of the shortest path of the model is calculated, and constraint conditions are set, thereby completing the intelligent logistics distribution path optimization modeling of agricultural products. Experimental parameters are set and traditional methods with path optimization modeling analysis methods are compared. From the comparison results, When the time is 10:00, the difference between the accuracy of the traditional method and the accuracy of the intelligent logistics distribution route optimization model is the largest, with a difference of 80%. It can be seen that the use of intelligent logistics distribution route optimization modeling and analysis has higher accuracy. It can be seen that the path optimization modeling and analysis method has higher precision in the analysis of agricultural products intelligent logistics distribution route, and provides an effective solution to ensure freshness of agricultural products.

Using adaptive weighted data fusion technology, the relative error of crop growth monitoring data is relatively large, accuracy is not high, and its fusion is not effective. In view of the above problems, the intelligent fusion technology of crop growth monitoring data under wireless sensor network is proposed. The technology consists of three parts: Using LEACH (Low Energy Adaptive Clustering Hierarchy) protocol to realize rapid processing and transmission of monitoring data. Accurate fusion of monitoring data through BP (Back Propagation) neural network; the two models are combined to construct the data fusion algorithm BPDFA (Back-Propagation Data Fusion Algorithm) model, so as to achieve intelligent fusion of crop growth monitoring data. By using the unique information processing characteristics of BP neural network, multi-information processing and transmission at the same time, the efficiency of processing is improved, and the fusion of crop growth information is realized. The results show that the intelligent fusion technology and adaptive weighted data fusion technology proposed in this study, the relative error is reduced by 3.72 °C, the accuracy is higher, and the fusion effect is better.

The development of deep learning in recent years has led to the development of natural language processing [1]. Question answering (QA) system is an important branch of natural language processing. It benefits from the application of neural networks and therefore its performance is constantly improving. The application of recurrent neural networks (RNN) and long short-term memory (LSTM) networks are more common in natural language processing. Inspired by the work of machine translation, this paper built an intelligent QA system based on the specific areas of the extension service. After analyzing the shortcomings of the RNN and the advantages of the LSTM network, we choose the bidirectional LSTM. In order to improve the performance, this paper add text similarity calculation in the QA system. At the end of the experiment, the convergence of the system and the accuracy of the answer to the question showed that the performance of the system is good.

The research on channel estimation technology is a core technology for mmWave massive MIMO in 5G wireless communications. This paper proposed a greedy iterative phase retrieval algorithm for channel estimation from received signal strength (RSS) feedback which is common in wireless communication systems and is used to compensate for temporal channels. We consider a Modified Gauss-Newton (MGN) algorithm to approximate the square term of the system model as a linear problem at each iteration and it is embedded in the 2-opt framework for iteration to get the optimal estimation. Our algorithm does not need to modify the system, but only need RSS feedback for channel estimation. The simulation results show that the algorithm performs better than the traditional conventional algorithm.

Aiming at the weakness of the human-computer interaction function in the current Web user display interface, long system feedback time and high error rate of data query, an interactive Web user interface display interactive design was proposed. Firstly, the requirements of user interface display are analyzed from the aspects of user role positioning, design availability analysis and interaction interface influence factors, and then the interactive scheme research of Web user interface is given. The overall interactive design scheme includes the operating program design, the usability design and the adaptive design of the interface color. From the above three points of view, the function of the design is expounded, and the convenience and friendliness of the interface are improved. Simulation results show that the proposed adaptive interaction design method can effectively shorten the feedback time of the system and reduce the error rate of data query. Adaptive display based on the advantages of human-machine interaction can better meet the needs of Web users.

Under cloud computing, traditional user browsing behavior data acquisition method cannot optimize data classification, which results in slow and low accuracy of data acquisition. For this reason, a fast method to obtain user browsing behavior data under cloud computing is proposed. Using node processing user browsing behavior data, complete the query the user browsing behavior data collection, provide the conditions for data classification optimization, the data to calculate the similar characteristics after multiple iterations data peak, peak according to complete the user browsing behavior data classification, the classification of output data integration, realize the cloud user browsing behavior fast data acquisition. Compared with the traditional data acquisition method, the data acquisition speed of the design method is increased by 20 min and the accuracy is increased by 45%. The experimental data show that the overall performance of the proposed method is better than the traditional method, and it has strong practicability and high reference value.

In this paper, kernel principle component analysis (KPCA) is employed to extract the features of multiple precipitation factors. The extracted principle components are considered as the characteristic vector of support vector machine (SVM) to build the SVM precipitation forecast model. We calculate the SVM parameters using particle swarm optimization (PSO) algorithm, and build the cooperative model of KPCA and the SVM with PSO to predict the precipitation in Guangxi province. The simulation results show that the prediction outcome, resulting from the combination of KPCA and the SVM with PSO, is consistent with the actual precipitation. Comparisons with other models also demonstrate that our model has advantages in fitting and generalizing in comparison other models.

Big data is a hot topic in the current academia and industry circles, which is influencing people’s daily lifestyles, work habits and ways of thinking. Due to the complexity of data itself and the huge amount of data, big data faces many problems in the process of collection, storage and use. It requires a new processing model to have greater decision making, insight and process optimization capabilities to accommodate massive, high growth rates and diverse information. The strategic significance of big data is not to master huge data information, but to conduct specialized analysis and processing of these meaningful data. This paper focuses on the analysis of IP network traffic under big data, and studies the sources of existing network traffic, the purpose of traffic analysis, and the common analysis methods for big data traffic. The structure and usability of Hadoop-based traffic analysis framework are mainly studied, and a new prospect is proposed for the future development direction.

Seam carving is a form of content-aware image modification. This modification can vary from resizing to clipping of content within an image. This can be easily used to alter images to achieve steganographic goals or the propagation of misleading information. Deep learning, particularly Convolutional Neural Networks have become prolific in today’s image-based intelligent systems. However, it has been found that convolutional networks specialized for image classification tend to perform poorly for steganalysis—specifically seam carving. In this paper, we propose a convolutional neural network architecture which is able to learn the nuances of seam carved images.

Transfer learning is a new topic in machine learning. Psychology holds that the process of learning knowledge from one to the other is a process of transfer learning. Transfer learning is different from machine learning which has to satisfy the following two conditions: (1) The training samples and testing samples must be in the same feature spaces. (2) There must be enough training samples to obtain an excellent training model. Because of the ability of transfer learning to solve problems with small samples and the ability to use historical auxiliary models to solve new problems, it is introduced in remote sensing image analysis. At first, this paper introduces some basic knowledge of transfer learning and enumerates some basic research examples. The research content of this paper mainly involves several problems based on transfer learning, such as target detection and recognition, image classification, etc.

Studying the paired approach of closely spaced parallel runways is of great significance for improving airport capacity and reducing flight delays, and has important theoretical and practical value. In order to study the longitudinal collision risk in the paired approach process, a kinematics equation is established to describe its motion process. Considering the influence of positional positioning error and aircraft wake motion, a longitudinal collision risk assessment model is established, and the calculation formula of relevant parameters in the model is given. Finally, the model is calculated by Matlab software, and the curve of collision risk with related parameters is given, and the rationality of the model is verified.

Firstly, by analyzing the original radar data of the aircraft in the airspace system, the historical operation information of each sector is extracted, and the traffic flow correlation between different routes of the same sector is considered. According to the characteristics of busy sector traffic flow data, a multi-dimensional data model of traffic flow with multiple related routes in the sector is constructed. Secondly, based on the data model, a traffic flow forecasting algorithm based on multi-time series machine learning is proposed. The main core idea of the algorithm is to use the time series clustering method to reduce the dimensionality of multi-dimensional traffic flow data, and then introduce the machine learning method for concurrent training. The training result obtains the optimal classifier group through competition. Finally, the multi-optimal machine learning integrated prediction method is designed to predict traffic flow. Taking the typical busy sector in China as an example, the proposed prediction method is verified. The research results show that the prediction results are better than the traditional single time series machine learning method, and the stability of the prediction results is good, which can fully reflect the dynamics and uncertainty of short-term traffic flow between sectors in each airspace, in line with the actual situation of air traffic.

Nowadays, in the complex electromagnetic environment, the detection of foreign satellite, the electronic interferences and the sensing data tampering in the process of consistent spectrum situation fusion and the electronic countermeasures reconnaissance and enforcement implemented by the enemy electronic attacks all pose serious threats to the communication performance of our electronic devices and communication systems. Therefore, how to detect these electromagnetic spectrum threats effectively is very important. The generative adversarial networks was applied in this paper, which is a method in deep learning, and an unsupervised solution for the above-mentioned electromagnetic spectrum threat signal prediction problem was provided, which has achieved good results. To carry out the detection experiments, three common electromagnetic spectrum threat scenarios were simulated. The prediction performance of the model is evaluated based on the prediction accuracy of the model. The experimental results have shown that the generative adversarial networks model used in this paper has a good predictive effect on the electromagnetic spectrum threat signals of a certain intensity.

Traditional methods of communication interference assessment belong third-party assessments that fail to meet the needs of real-time assessments. This paper proposes an interference level evaluation method under the nonlinear classification algorithm. Firstly, building data set with the eigenvalues that affect the interference effect, and then simulation verify by BP neural network and support vector machine. The simulation results verify the feasibility in communication interference assessment and providing the possibility for real-time evaluation.

Traditional methods of improving wireless network security are through software-level device identification, such as IP or MAC addresses. However, these identifiers can be easily changed by software, making wireless network communication a high risk. In response to these risks, radio frequency fingerprinting technology has been proposed. Since the radio frequency fingerprint is an essential feature of the physical layer of the wireless communication device and is difficult to be tampered with, it is widely used to improve the security of the wireless network. Based on the physical layer characteristics of the communication system, this paper has established a relatively complete RF fingerprint identification system to realize the identification and classification of the devices. Two signal starting point detection methods and two RF fingerprint feature extraction methods are studied in this paper. The detailed results are obtained by combining the dimensionality reduction and classification methods. Finally, an optimal identification scheme was found to achieve a classification accuracy of more than 90% when the signal-to-noise ratio is greater than 15 dB.

The research of radio-frequency (RF) power amplifier model has always been one of the most important breakthroughs in the emitter feature extraction and specific emitter identification. Through the establishment of RF power amplifier model, we can extract feature parameters of the specific emitter. In this paper, we discuss the research issues of the specific emitter feature extraction and individual identification based on RF power amplifier nonlinear model, summarize the nonlinear distortion and modeling method of the power amplifier. Furthermore, we analyzed the applicability of these models.

With the increasing integration of the control system and the continuous development of the PCB processing technology, the testing of PCB becomes more and more difficult. The manual detection of PCB fault points can no longer meet the needs of the industry. In this paper, a universal circuit board detection system based on image automatic positioning is proposed. The system can accurately collect the coordinates of the measured points by using the image information, and record the edge information of the circuit board. The position error caused by the fixed position of the circuit board is corrected in real time. At the same time, the automatic correction function of the pressure drop of the system can automatically correct the voltage drop produced by the hardware circuit and improve the accuracy of the measuring system. The experiment proved that the fault point detection system based on image location can effectively improve the efficiency, accuracy and universal of circuit board detection.

Modulation recognition is an important part in the field of communication signal processing. In recent years, with the development of modulation recognition technology, various problems have emerged. In this title, we propose an improved recognition framework based on SVM, which extracts the entropy feature of the signal and distinguishes it from the traditional modulation recognition framework. We combine the training set with the test set first, then carry on the density clustering to the whole data set. The data set after the cluster is extracted according to a certain proportion to build a new training set, and the new training set is used to train the SVM. Finally, the data of the test set is modulated by the modulation recognition. Experimental results show that the proposed method improves the recognition rate of traditional SVM framework and enhances the stability of traditional SVM framework.

The sparse target location algorithm based on orth can solve the problem that the sampling dictionary does not satisfy the RIP property. Compared with the traditional method, the orth preprocessing can reduce the energy consumption and communication overhead, but the orth pretreatment will affect the sparsity of the original signal. So that the positioning accuracy is affected to a certain extent. In this paper, a sparse target location algorithm based on QR-decomposition is proposed. On the basis of orth algorithm, the sampling dictionary is decomposed by QR, which can’t change the sparsity of the original signal under the premise of satisfying the RIP property. The problem of sparse target location based on network is transformed into the problem of target location based on compressed perception, and the localization error is reduced. The experimental results show that the location performance of sparse target location algorithm based on QR-decomposition and centroid algorithm is much better than that the sparse target location algorithm based on orth, and the accuracy of target location is greatly improved.

As an important text type, news texts have great research value in data mining, Such as hotspot tracking, public opinion analysis and other fields. News text clustering is a common method for studying the trend of news and hotspot tracking. Most of the existing clustering methods are based on the vector space model, with calculating the TF-IDF of words in the news text as feature items of the text. To improve the performance of clustering in the news texts, this paper presents a new clustering algorithm, this algorithm expresses the news text as a series of Text labels, which effectively solves the problem that the data latitude is too high, and the clusters is too hard to express. At the same time, by using a conceptual clustering algorithm, this method effectively reduces the number of comparisons. The experimental results show that the algorithm based on similarity of text labels improves the quality of clustering compared to traditional clustering methods.

Surveillance and conference videos have become increasingly important in our daily life, which brings a huge amount of video data. Existing coding standards were originally designed for generic video contents. The backgrounds are generally static in the surveillance and conference videos. The background coding errors will propagate to the subsequent frames in coding the videos. In this paper, a background error propagation (BEP) model based Rate Distortion Optimization (RDO) scheme in HEVC is proposed for the surveillance and conference videos. Firstly, the global RDO scheme is proposed to efficiently exploit the background error propagation. Secondly, a BEP model is studied to express the linear relationship between the distortion of the first frame and that of its subsequent frames. Based on the BEP model, enhanced frames are proposed to be coded with a small quantization parameter (QP) offset so as to improve the global performance. Thirdly, a bi-exponential decay model is proposed to investigate the variation of the error propagation ratio as the frame order increased. Based on the decay model, a periodical optimization scheme is presented by deploying the enhanced frames periodically. Experimental results show that the proposed algorithm achieves 11.15% bit-rate reductions on average under the low delay condition.

In order to improve outdoor performance of vehicle binocular system, the stereo matching algorithm based on “3bit-Census Transformation & An Adaptive window aggregation based on edge truncation & Fast Parallax Calculation” was proposed. The stereo matching algorithm based on this framework improved the robustness, matching accuracy and efficiency of the calculation from different stages. The experimental results show that the algorithm proposed in this paper is better than the traditional algorithm and can meet the requirements of the vehicle binocular system.

The civil aviation safety management system requires accurate prediction of the future safety status, but there are often many uncertain factors in the occurrence of air traffic insecurity events. In order to study its development trend and strengthen the accurate analysis and prediction of unsafe events, a combined forecasting model based on Markov correction process is proposed. Firstly, apply the grey system theory to construct a GM (1, 1) model. Then, based on the grey prediction model and the exponential smoothing method, combination forecasting model is established. And according to the standard deviation of the prediction result, the weight is determined to correct the data. Finally, combined with the Markov method, the probability transfer matrix is determined and the results are optimized. Based on the statistics of civil aviation insecure events in the past ten years, the prediction accuracy of the optimized model is significantly higher than that of the single gray prediction model or the exponential smoothing prediction model, which verifies the effectiveness of the method.

In this paper, analysis is conducted on the risk assessment regarding the vertical collision of CSPR (Closely Spaced Parallel Runways) paired-approach, to ensure flight safety. A vertical kinematics equation is established with analysis of CSPR paired approach and starting from the preconditions that the proceeding aircraft altitude is lower than that of the following aircraft during paired approach: the time consuming of passing initial safety separation by the proceeding aircraft decelerated less or greater than that of the proceeding aircraft with uniform speed. Based on the two conditions, its corresponding risk-evaluation model is established, proceeding from the aircraft ADS-B data and the analysis on the relation between aircraft position error and altitude maintain ability, relevant model parameters specified. Conclusion has been achieved on risk assessment that implementing vertical collision risk of paired approach has little to do with aircraft type and initial longitudinal separation, but has more correlation with initial vertical interval and aircraft altitude maintain ability; rules of at least 180-m vertical interval and altitude error not exceeding 40.77 m (within 95% flight time) must be obeyed when paired approach applied.

The rules issued by third-party applications may have direct violations or indirect violations with existing security flow rules in the SDN (software-defined network), thereby leading to the failure of security rules. Currently, existing methods cannot detect the rule collision in a comprehensive and fine-grained manner. This paper proposes a deep detection mechanism for rule collision that can detect grammatical errors in the flow rules themselves, and can also detect direct and indirect rule collisions between third-party and security applications based on the set intersection method. In addition, our mechanism can effectively and automatically resolve the rule collision. Finally, we implement the detection mechanism in the RYU controller, and use Mininet to evaluate the function and performance. The results show that the mechanism proposed in this paper can accurately detect the static, dynamic and dependency collisions of flow rules, and ensure that the decline of throughput of the northbound interface of the SDN network is controlled at 20%.

In light of the existence of the software defined networking (SDN) southbound communication protocol OpenFlow, and manufacturers’ neglect of network security, in this paper, we propose a protection scheme for encryption at the source of the communication data that is based on the Kerberos authentication protocol. This scheme not only completes the identity authentication of and session key assignment for the communication parties on an insecure channel but also employs an efficient AES symmetric encryption algorithm to ensure that messages always exist in the form of ciphertext before they reach the end point and thus obtain end-to-end security protection of communication data. At the end of this paper, we present our experimental results in the form of a forwarding agent. After that, the performance of the Floodlight controller is tested using a CBench testing tool. Our results indicate that the proposed source encryption scheme provides end-to-end encryption of communication data. Although the communication latency increases by approximately 12% when both transport layer security (TLS) and source-encrypted are enabled, the source-encrypted part of the increase is only approximately 4%.

With the development of network technology and multimedia technology, digital video is widely used in news, business, finance, and even appear in court as evidence. However, digital video editing software makes it easier to tamper with video. Digital video tamper detection has become a problem that video evidence must solve. Aiming at the common inter-frame tampering in video tampering, a tampered video detection method based on structural similarity mean value and support vector machine is proposed. First, the structural similarity mean value feature of the video to be detected is extracted, which has good classification characteristics for the original video and the tampered video. Then, the structural similarity mean value is input to the support vector machine, and the tampered video detection is implemented by using the good non-linear classification ability of the support vector machine. The comparison simulation results show that the detection performance of this method for tampered video is better than that based on optical flow characteristics.

Nowadays, RFID positioning has become the preferred technology in indoor positioning because of its strong anti-interference ability, short recognition time, large amount of storage data and low cost. In this paper, based on RFID technology, a method of virtual tag is proposed to further optimize the adjacent area, the positioning accuracy is further improved without increasing extra cost and signal interference which has more superior performance and a higher practical value, so as to achieve the purpose of optimization.

The traditional drug detection data management system has the disadvantages of limited submenu generation and uneven distribution of management rights. In order to solve these problems, a new data management system based on big data is designed. Through the two steps of .NET framework and B/S detection module design, the hardware operation environment of the new system is completed. On this basis, determine the MyEclipse node and the detection process. Under this precondition, all the process parameters related to drug data are stored in the system database for a long time, and the total amount of E-R data can be determined, and then the design of drug testing data management system can be completed. The experimental results show that compared with the traditional system, the management authority distribution uniformity of the system can reach 81.57%, which is much higher than that of the traditional method. The application of the new system can effectively improve the sub-menu generation rate.

In the original version of the book, the following belated correction has been incorporated: The author’s name was corrected from “Jingmeifang Li” to “Jingmei Li”.