Abstract:The knowledge graph has been proposed by Google as a knowledge base to enhance its search function, and has been rapidly developed in recent years. As the value of knowledge graphs is constantly being explored, many domain knowledge graphs are rapidly being built. This study clarifies the definition of domain knowledge graphs by comparing domain knowledge graphs with general knowledge graphs, introduces the architecture of domain knowledge graphs, and uses medical knowledge graphs as an example to explain the construction techniques of knowledge graphs. Finally, this study introduces the development of the current popular domain knowledge graphs, and provides a comprehensive summary of the current domain knowledge graph status.

Abstract:Retinex algorithm is an important method in the field of color image enhancement, and it has rich connotation. Aiming at the limitations of traditional Retinex algorithm in color distortion and halo, the paper discusses and analyzes a series of related improved methods based on color space transformation, color correction, and estimated illumination component design. Finally, the further possible research directions of Retinex algorithm are pointed out.

Abstract:This study built an L-system model to dynamically simulate the growth and visualize the three-dimensional structures of the rice root system using the L-studio modelling environment. Firstly, the rice root structures and characteristics in the growth processes were analyzed and several modules representing the tip of root, node of branching, node to generate the seminal root, and nodes to generate the crown roots were defined. Then, a set of production rules were developed covering the growth and branching of a single root, generation of the seminal root, and crown roots. In the model, the directions of the single root depended on two factors, a gravity factor G and a deflection factor D. They were introduced into the rice root growth production rule to investigate their influences on the growth process of the rice root system. Finally, the axiom including the notes where the seminal roots and crown root were generated was defined. With the iterative replacement of the defined modules and interpretation of turtle modules in the current strings, the continuous growth processes of the rice root system were generated. In the experiments, with the fixed parameters for the roots in different orders, the rice root system was successfully simulated and visualized with LPFG and L-studio. The experimental results demonstrated that gravity factor G determine the degree the roots bend toward the earth while the deflection factor D determine the degree the roots deflect around y axis randomly and rice root growth simulation at G≈0.3 and D≈0.3 is more realistic. This study will benefit further research on the simulation and visualization of other plant root.

Abstract:At present, the defects on the surface of the light guide plate are mainly detected by human eye, only a few manufacturers use the traditional image processing methods. In the imaging of high-resolution industrial cameras, the defects of the light guide plate are still extremely small, the characteristics of different defects are different, and the light guide points of the entire light guide plate are densely distributed and uneven,which leads that the traditional image processing detection methods require experienced visual experts to carry out a large number of feature extraction algorithm programming work and expensive code maintenance cost, low accuracy and poor stability. Therefore, a defect detection method based on deep learning semantic segmentation is proposed. This method can learn and extract the characteristics of the light guide plate defects by training the neural network to avoid the complicated feature extraction algorithm programming. First, the collected light guide plate defects are marked for making a sample set. Secondly, the pre-trained Pyramid Scene Parsing Network (PSPNet) is used to retrain the labeled samples using transfer learning. Further, the trained model is used to achieve detection of defects of the light guide plate. Since the separate deep learning semantic segmentation defect detection method usually cannot meet the industrial practical application requirements, it is necessary to combine the simple machine vision method to make a second judgment and screening of all suspected defect regions detected by the deep learning semantic segmentation method. The experimental results show that the detection rate of the three defects of bright spots, dark spots, and scratches is as high as 96%, which can basically meet the industrial testing requirements.

Abstract:In order to further improve the accuracy of network anomaly detection, based on the analysis of existing intrusion detection methods, this study proposes a network packets intrusion detection method based on Convolutional Neural Networks (CNN) and Support Vector Machine (SVM). The method first preprocesses the data into a two-axis matrix. In order to prevent the algorithm model from over-fitting, the permutation function is used to randomly shuffle the data, and then the CNN is used to learn the effective features from the pre-processed data. Finally, this method uses SVM classifier to classify the vectors. In the dataset selection, we use the authoritative dataset commonly used in network intrusion detection—Kyoto University honeypot system dataset. This method proposed in this study is compared with the existing models with high detection rates, such as GRU-Softmax and GRU-SVM. The model has improved the highest accuracy by 19.39% and 12.83% respectively, which further improves the accuracy of network anomaly detection. At the same time, the method has greatly improved the training speed and test speed.

Abstract:The traditional sign language recognition only relies on the underlying features selected manually, which is difficult to adapt to the diversity of sign language image background. A method of sign language recognition based on the multi-factor skin color segmentation and the improved VGG network is proposed in the study. The collected sign language images are initially segmented by an elliptic model. The skin color region is excluded according to the maximum connected domain, and the skin color regions outside the hand region is removed by centroid positioning method, so as to realize the accurate segmentation of sign language images. The VGG network is improved by reducing the number of convolution and full connection, which reduces the required storage capacity and the number of parameters. The gray-scale image of the segmented sign language is taken as the input of the network, and the improved VGG network is used to establish the recognition model of sign language. By comparing the different structure of the network model of sign language recognition rate of the image, show that the improved VGG networks can effectively study characteristics, the average image sign language recognition rate is above 97%.

Abstract:The author reviews the related literatures in recent decades and summarizes existing research results from the definition of modeling and simulation implementation quality and related concepts, simulation work, simulation data, modeling process, model, results, etc. Most of current research partly foucuses on simulation work and data. The system research from the whole life cycle is relatively few, and there is a lack of a standardized framework to ensure the final effect of modeling and simulation implementation. However, the research from the perspective of the whole life cycle is more comprehensive, so it is very necessary. Finally, the research direction of modeling and simulation implementation quality are forecasted from the aspect of life cycle.

Abstract:During transmission of streaming video data, there are quite high requirements for timeliness, transmission efficiency, and transmission accuracy. This study focuses on the application of real-time big data stream processing, analyzes the difficulties and key technologies required for streaming media video data in real-time transmission, and achieves a high-performance, low-latency distributed real-time transmission system by the Storm processing technology. Then, we set up the Storm framework on Linux system, design and implement the transmission topology task of streaming video data, while deploy Zookeeper to provide efficient and reliable distributed coordination services for this framework, and finally build a streaming server for video storage and client-on-demand. After setting up the target system, it passed the real-time transmission test of large-scale streaming media data.

Abstract:As the existing micro-credential systems rely on centralized management model, the security of credentials cannot be guaranteed. To tackle with such problem, this study introduces blockchain technology into the architecture design of micro-credential systems, utilizing its distributed structure and technical characteristics (i.e., transparency and immutability). This study first illustrates the drawbacks of existing micro-credential systems, and introduces the basic concept of Blockchain ecosystem. Secondly, an architecture of Blockchain-based micro-credential system is designed and proposed in this research, to provide efficient and intuitive services related to micro-credentials. Finally, a prototype of the proposed system architecture is implemented, and evaluated in terms of performance and security. The experiment results show that the proposed approach in this study can be applied to Blockchain-based micro-credential system, to facilitate its design and implementation.

Abstract:Because the current diagnosis and treatment decision support system adopts a single subject decision-making method, resulting in low diagnosis and treatment accuracy and low accuracy of the obtained data classification results, a multi-disciplinary collaborative diagnosis and treatment decision support system based on improved K-Nearest Neighbor (K-NN) classification algorithm and Support Vector Mechine (SVM) is proposed and designed. Based on the overall framework of the system, the database system module, human-computer interaction module, and diagnosis and treatment reasoning module are designed. The diagnosis and treatment reasoning module is the software core of the system. The reasoning engine is established by improved K-NN classification algorithm and SVM. With the help of computer, medical cases similar to the patient’s disease information are searched, and similarity matching is carried out. According to the matching results, a new clinical case is constructed based on patient symptom set. The concept of Clinical Document Architecture (CDA) is introduced to realize the effective fusion of improved K-NN classification algorithm and SVM algorithm, and to complete the multi-disciplinary collaborative diagnosis and treatment decision. The experimental results show that, compared with the traditional system, the system has high accuracy in diagnosis and treatment decision-making, the average value of the evaluation index is 95.98%, and the accuracy rate of classification results is high. With the help of the system, it can improve the diagnosis accuracy of doctors and reduce the misdiagnosis rate, and the operation complexity is low.

Abstract:China is a traditional agricultural country. Agriculture is not only the foundation for the construction and development of the national economy, but also the guarantee for the normal, stable, and orderly operation of the society. However, the annual losses due to crop pests and diseases are huge, and the traditional methods for identifying crop pests and diseases are not ideal. At the same time, the rapid development of deep learning in recent years has made great progress in image classification and recognition. Therefore, this study constructs an image recognition model for crop pests and diseases through deep learning based methods, and improves the convolution network loss function for sample imbalance problems. The model can effectively identify crop pests and diseases and the accuracy of the model is further improved after optimizing the loss function.

Abstract:With the emerging large image sets and the rapid development of computer hardware, especially GPU, the deployment of Convolutional Neural Network (CNN) model on embedded devices with limited computing resources becomes a challenging problem. Overheating of power equipment can be identified from infrared thermal images. Because of the fading of infrared radiation in the air, the result of infrared temperature measurement is lower than the actual value. In this study, an efficient CNN based on embedded devices is proposed for thermal fault detection of power equipment. The backbone network of SSD algorithm is replaced by MobileNet. At the same time, batch normalization is combined with the previous volume to reduce model parameters, improve reasoning speed, and make it run on a lightweight computing platform. To solve the problem of infrared radiation loss in the air, an infrared temperature correction unit based on BP neural network is proposed. Based on the above innovation, a thermal fault detection system for power equipment is designed. Experiments and field applications show that the proposed method has high accuracy and reasoning speed.

Abstract:Urban traffic congestion is seriously harmful, directly causing time delays, energy waste, emission increase, and the living standards reduction of residents. At current cross-road intersection, traffic lights switching time is relatively fixed, the actual situation of traffic jams at intersections occur in bad weather or in the event of a traffic accident. Therefore, this study proposes an adaptive algorithm for multi-directional red light time to ease the traffic congestion at intersections. Using video image processing algorithm to judge road traffic congestion, we set the traffic light time according to road conditions, and design the corresponding control system. Simulation results demonstrate that the traffic efficiency by this proposed adaptive algorithm is higher than that of the traditional traffic light running time control method during the heavy traffic.

Abstract:A new intelligent adaptive adjustment reading shelf is designed in this study, and the application of dual CPU is adopted. In this system, the raspberry processor ARM Cortex-A53 is used to design the system. The whole system is composed of three parts: the reading shelf terminal, the cloud server, and the application end user platform. The terminal of the reading shelf is equipped with a single camera, which can recognize the reader’s forward, backward, left, right, head, bow, and correct sitting posture in real time, and can control the reading shelf up and down, right and left adjustment through controlling the rotation of the stepping motor, so as to track the reader’s posture automatically. In addition, the intelligent reading shelf can upload the users’ time, reading attitude, and other information to the cloud server and form the user archives for the users to visit. The design product can be widely applied to kindergartens, primary and secondary schools, library reading rooms, and ordinary families. It helps readers to sit upright, develop good reading habits, and achieve significant social and economic benefits.

Abstract:Aiming at the problem that using multi-source heterogeneous environmental data is difficult, based on the general knowledge graph, this study constructed the environmental knowledge graph by fusing various environmental data. Firstly, the environmental data are obtained by using Web crawlers and preprocessed later; then, data conversion, text extraction, and data fusion technologies are used to build the knowledge graph; finally, the generated knowledge graph is stored in the graph database, and then we build a knowledge graph application platform to provide a way to searching information recursively and to realize the visualization environmental knowledge graph. Such platform is more comprehensive thus it may offer better references for the relative personnel.

Abstract:The access control management methods for important places such as military weapons warehouses are insufficient in security. In order to solve the defects, we design an access control system based on improved convolutional neural network. This paper first introduces the basic knowledge of convolutional neural networks, then introduces Particle Swarm Optimization (PSO) algorithm to design and optimize initial weights and thresholds of convolutional neural networks. After designing, experiment with the MNIST handwritten digital dataset is carried out. The results demonstrate that the modified convolutional neural network can make the convergence speed faster, and the loss is smaller, so the outcome is obviously better than the traditional convolutional neural network. On this basis, according to the actual working conditions of the troops, PSO is applied in the MTCNN and SIAM-ResNet face detection algorithm, the access control system based on improved convolutional neural network is designed, which makes the access control of important places in the army have higher security and reliability.

Abstract:The main functions of a Battery Management System (BMS) are to monitor and to manage the normal work of each single battery cell in the power battery pack. The information of the battery are sent out by BMS in the form of the CAN message. By collecting and analyzing the BMS messages, the performance of the battery can be further analyzed and evaluated. This study used Raspberry PI as the carrier, realizing the CAN message collection by designing an extension board of the Raspberry PI, and the messages were interpreted as CAN protocol DBC file format in Raspberry PI. According to the open circuit voltage of all cells in the battery pack, the battery consistency was quickly evaluated.

Abstract:In this study, the convolutional neural network under the deep learning framework is applied to the field of cervical cell identification to achieve automatic classification of cervical cell images. Firstly, the cervical cells are pretreated, and the problem of different image input sizes is solved by nuclear cutting, the image is flipped and translated, the data set is expanded, and the sample size imbalance is solved. Then the VGG-16 network is selected for improvement. The improved VGG-16 network is used for feature extraction and cell classification. The migration learning method is used for network pre-training, which speeds up the network convergence speed and improves the classification accuracy. Finally, through the training of the network, it achieves better result. According to the classification results, the classification accuracy is improved compared with the manual extraction feature design classifier. The accuracy of two categories classification is 97.3%, and the accuracy of the seven categories classification is 89%. The experimental results show that the convolutional neural network automatically classifies the cervical cell images, and the classification accuracy is better than that of the artificial extraction feature classifier, and the classification results are not affected by the segmentation image accuracy.

Abstract:Community structure is one of most important characteristics of complex networks. The community detection problem based on modularity is NP-hard as a combinatorial optimization problem, which is often solved by heuristic algorithms. Jaya algorithm is a simple and effective meta-heuristic method for solving continuous optimization problems. In this study, the strategy of discreting Jaya algorithm for complex network community discovery is given on the basis of updating the population individuals according to the way Jaya algorithm works, that is, an individual is updated close to the best solution and far away from the worst solution, and thus a discrete Jaya algorithm for complex network community discovery is proposed. Experiments show that the proposed algorithm has the advantages of high resolution and automatic determination of the number of communities compared with the classical algorithms in several real network instances and a class of artificial network instances.

Abstract:The small-sized traffic signs actually detected by the YOLOv2 algorithm are of poor quality, low recognition rate, and poor real-time performance. This study proposes a traffic sign detection method based on improved YOLOv2. Firstly, the image is enhanced by histogram equalization and BM3D method, with high-quality images. Moreover, the top-level convolutional layer output feature map of the network is finely divided to obtain fine-grained feature maps to detect high-quality, small-sized traffic signs. Finally, the loss function is improved by normalization and optimization of the confidence score ratio method. Experiments were carried out on a new data set combining CCTSD (China Traffic Sign Detection Dataset) and TT100K dataset. Compared with the YOLOv2 network model, the network recognition rate increases by 8.7% and the recognition speed of the model is improved by 15 FPS. Experimental results show that small-sized traffic signs can be accurately detected by proposed method.

Abstract:As one of the classic research directions in the field of natural language processing, the task of target-specific sentiment analysis is to determine the sentiment polarity of a specific target based on contexts. The key to improve the performance of this task is how to better mine the semantic representation of specific target and contexts. This study proposes a multi-attention network with phrase features. By introducing phrase-level semantic features, a multi-attention network with multi-granularity features is constructed to improve the expression ability of the model effectively. The experimental results on the SemEval2014 Task4 Laptop and Restaurant datasets show that the PEMAN model proposed in this study has a certain improvement in accuracy compared with the benchmark model.

Abstract:To solve the polysemy troublesome problem of tourism text in feature extraction, a Chinese scenic spot named entity recognition model based on fusion language model is studied for the problem of attraction alias in the visual recognition of tourist travel texts. Firstly, the BERT is used for tourism text feature extraction to obtain the word granularity vector matrix. BiLSTM is used to extract the context information. The CRF is used to obtain the global optimal sequence, and finally the tourist attraction entity is obtained. Experiments show that the performance of the proposed model is significantly improved. In the test of scenic spot identification in the actual tourism field, compared with the previous research, precision and recall rates are increased by 8.33% and 1.71%, respectively.

Abstract:This study proposes an algorithm for semantic segmentation of targets in images based on fully convolutional neural networks and a new method to make and augment dataset. The algorithm primarily segments the targets from images using improved fully convolutional neural networks, OTSU method is applied to binarize images and segment the general areas of targets, finally, the fully connected conditional random field algorithm is used to correct the general areas of targets and get the final results. This algorithm achieves the accuracy of 85.7% and speed of 0.181 second per image on test set, and prepares for further analysis of targets in images.

Abstract:Chinese word segmentation is an important task in Chinese natural language processing. One of bottleneck problems in Chinese word segmentation is Out-Of-Vocabulary (OOV) detection. This study proposes an unsupervised OOV detection method based on improved PMI algorithm and minimum branch entropy combining strategy. Firstly, the punctuation marks and special characters which are not related in the text are removed. The improved PMI algorithm recognizes the string with strong cohesion in the text, and gets the candidate OOV through the filtering of the stop word list and the core vocabulary. Then the minimum branch entropy of candidate OOV is calculated, when the term frequency-minimum branch entropy threshold is met, the output is the OOV. Through theoretical and experimental analysis, the algorithm can generate a personalized OOV dictionary for different texts, and does not require long-term learning and training to adjust parameters, and has a certain improvement in the accuracy and recall rate of detection.

Abstract:With the wide application of the mobile Internet, a series of mobile Internet applications such as the smart community have received much more attention for citizens, especially the anti-fall detection of the elderly who are at home. In view of the fact that some elderes fall down occasionally without timely detection and alarm, which can not be aided in time, resulting in more serious safety problems, this study proposes a fall detection method. In this method, it first scans a specific human body, constructs a human body model using poser, and then maps the two-dimensional coordinates to the corresponding three-dimensional coordinates according to the position of the joint points during the motion and uses the spatial position error prediction algorithm to perform joint points on the mapped three-dimensional coordinates, then aggregates the predicted sub-joints into the three-dimensional space axis of the parent class and predicts the motion state of the parent joint point. When the child joint point and the parent joint point prediction result are simultaneously in a falling state, the proofed result is falling state. Since the established motion model has higher realism in motion characteristics, the data changes of the joint points are real and reliable. After having done experiments with a large number of experimental data, it is proved that this method can accurately and real-timely detect the reaction state when the elder falls down, and the detection accuracy is 99%. Therefore, this proposed method is effective and reliable for fall detection.

Abstract:This paper presents a method based on MPU9250 microprocessor for human motion recognition. The user performs various types of sports while wearing the bracelet, and the bracelet automatically collects and stores the acceleration data generated by the user during the movement. Analysis of these data can identify the type of human motion. The acceleration of the motion in the X, Y, and Z directions is collected by the acceleration sensor embedded in the wristband. After filtering by the filtering algorithm, the data is analyzed in the time domain and the frequency domain respectively, and then subjected to feature engineering extraction. There are 34 related features, feature selection algorithm is used to select the 16 main features, reducing the complexity of the algorithm. Experiments compare the three methods, namely Support Vector Machine (SVM), decision tree, and random forest, in classification of five sports, modes such as walking, running, badminton forehand swing, table tennis, and rowing. The results show that the random forest has the best accuracy which can reach more than 97%.

Abstract:The traditional recommendation method based on collaborative filtering can mine the hidden features in the score, but the recommendation process takes a long time, and the score matrix is sparse, resulting in a large error between the real value of the sample and the predicted value. Neural network can quickly calculate the characteristics of objects through batch training. The number of parameters is significantly reduced by the local perception and parameter sharing of convolutional neural network. The calculation time can be shortened by using common neural network and convolutional neural network to realize recommendation together. By adjusting the parameters of neural network, the reasonable feature vector and convolution kernel size for convolutional neural network can be designed to improve recommendation speed and accuracy. Experimental results show that the method of combining neural network with convolutional neural network can reduce the mean value of absolute error of recommendation to below 0.67, and greatly improve the accuracy and effectiveness of recommendation.

Abstract:Aiming at the target person’s emotional changes, this study proposes a method of emotion prediction to identify, predict, and analyze emotions. Before sentiment prediction, a sentiment quantitative algorithm is used to normalize the sentiment data set to obtain the degree coefficient corresponding to each sentiment, which lays the foundation for the next sentiment prediction. Then, we summarize the mood changes of the target person for one day to get a main mood, and then use the mood prediction algorithm to get the final prediction result. In this study, Bidirectional Encoder Representations from Transformers (BERT) neural network is used to model the emotion of short dialogues in order to achieve real-time emotion prediction of target person. The results show that the application of the training model in this study can effectively determine the future mood fluctuations of the target person.

Abstract:Object-oriented programming is increasingly pursuing the reusability and flexibility of the program. It is difficult for programmers with less experience to obtain reusable and flexible programs directly. The software design pattern is to extract the experience of object-oriented programmers, and then to summarize them. In the template method pattern, the parent class defines an algorithm framework, uses the template method to specify the execution steps of the algorithm, and delays the variable steps to the subclass implementation. Each different implementation needs to define a new subclass. The system will be larger with maintainability and readability. Therefore, embedding the abstract factory pattern into the template method pattern forms a composite pattern. The core of the composite pattern design is to provide an interface for creating objects for each variable step that is delayed to the subclass. The interface is for a complete definition of product family. The composite mode not only ensures the stability of the algorithm structure, but also separates the specific implementation class, which enhances the robustness, reusability and flexibility of the program.

Abstract:In Chinese essay, compound sentences are the majority. Recognition of relation category is screening for semantic relation of clauses in a compound sentence, and it is the key to analyze the meaning of the whole compound sentences. In a non-saturated compound sentence, the relation words are absent. So, the non-saturated compound sentence can not be classified by the features of the relation word collocation. In this work, an unbalanced corpus of non-saturated compound sentences with two clauses is taken as the research object. This study proposes a convolutional neural network for relation classification that automatically learns features from two clauses and minimizes the dependence on pre-existing natural language processing tools and language rules. The model fuses the features of relation to improve the performance. The experimental results show that the accuracy is 97% and that the proposed model outperforms the best baseline systems with sentence level features.

Abstract:The study researches indoor thermal comfort from the perspective of smart home, analyzes the thermal comfort evaluation method of PMV, and points out that some of its parameters are difficult to obtain in the smart home scene. The study proposes to introduce the climatic and environmental characteristics to fit the PMV formula while ignoring wind speed and average radiant temperature. The research uses BP neural network algorithm optimized by Differential Evolution (DE-BP) to establish a fitting model, DE algorithm optimizes parameters of neural network, neural network training uses momentum-accelerated stochastic gradient descent algorithm, and adds the normalization layer and L2 regularization of the affine transformation. The test results show that the model is better than the traditional BP neural network in terms of convergence speed, stability, and generalization performance, and can be used within a small error range. It is applied to the system for calculating thermal comfort and reduces the difficulty of input parameters.

Abstract:At present, the demand for large-scale scene model generation is increasing. A region-based automatic Levels Of Details (LOD) construction algorithm is proposed. The algorithm is based on the dynamic mesh simplification algorithm. In the process of game design and production, game developers often need to work on the art. The model provided by the department staff is optimized in order to simplify the number of faces of the model and not change the appearance of the model. Today’s popular LOD technology is very good at handling this situation, judging the distance between the model and the camera. After a certain range is exceeded, the models of different levels are automatically retrieved. When the model is far away from the camera, the model with a low number of faces is replaced by the high-modulus. This can increase the frame rate and reduce the number of triangles and the number of vertices in front of the camera to reduce rendering stress. The simplification of the general model is divided into the simplification of static and dynamic models. In most cases, the programmer will let the art department provide several sets of different face models or reduce the face of the high model and save it into multiple mesh through the model simplification tool. When the program is running, according to the distance between the model and the camera, the mesh is replaced dynamically. This is a static method. Here we will try to use a combination of dynamic mesh simplification and LOD technology. This new algorithm greatly simplifies the operation process, while the artist only needs to provide a model, then the programmer can use this method to generate low-modules of different magnitudes, and automatically pick up models of different precision according to the distance between the camera and the model.

Abstract:To solve the job-shop scheduling optimization problem with robotic cell, several jobs with specific processing operations can be processed on several processing machines, and the handling robot can carry the jobs between the loading/unloading stations and the processing machines. In the real world, due to the uncertainty, especially the processing unit with inventory, the completion time of the job is required in a time window, rather than a specific time point. Therefore, considering the complexity and constraints of the job-shop with robotic cell, the objective is to minimize the total weighted earliness and tardiness. An improved meta-heuristic algorithm is proposed, which combines memetic algorithm with local search technology (variable neighborhood descent). The optimal job processing order and robot handling sequence can be obtained simultaneously. Computational experiments show that the proposed algorithm is more efficient than other algorithms.

Abstract:In order to solve the problem of inaccurate target location caused by camera jitter and rotation, a novel fast anti-rotation matching algorithm is proposed, which uses the convolution operation mode formed by the operation to reduce the matching time exponentially. A large number of experimental results show that using this algorithm, the target automatic calibration of digital camera visibility instrument is accurate, the average time of single calibration is 24.2 ms, which is at least 200 times faster than the original algorithm in the same situation, and the observation accuracy is more than 7 times higher than the ordinary algorithm, nearly 5 times higher than the manual positioning observation. The measurement results of the digital camera visibility meter based on the algorithm meet the requirements of the World Meteorological Organization for the development of visibility meter, and the price is lower, so it has a broad application prospect.

Abstract:Aiming at the problem of poor information dumping in the remote online fault analysis of a turbine vibration signal, we use the current popular cloud database technology to realize information dumping from power plants to expert diagnostic centers. Combining the software architecture and implementation technology of existing acquisition systems, we propose a solution for information dump using cloud database, then mainly consider the performance guarantee of vibration information dump, including the integrity, timeliness, and safety of information. The corresponding technical means are used to ensure these indicators. Laboratory validation tests show that the vibration information of the acquisition system can be effectively dumped to the database of the diagnostic center, and the relevant performance indicators can also be effectively guaranteed.

Abstract:Based on the current high-voltage transmission lines, towers ledger information, inspection information presented island state, this article illustrates an application case that multidimensional data of high-voltage transmission lines. Aiming at the digital information, document information, picture information, and video information generated by the defective description of helicopters, UAV, and manual inspections, the platform technology is proposed with focusing on the goal of ensuring the safety of the transmission line operation.

Abstract:The air quality is closely related to people’s lives. The prediction results of air quality are the basis for air quality control. Therefore, how to improve the prediction accuracy of air quality is the focus of this study. The Community Multiscale Air Quality modeling system (CMAQ) and the Comprehensive Air quality Model with extensions (CAMx) are two commonly used numerical models of air quality. The prediction principles are based on atmospheric physical and chemical methods to simulate the process of pollutant transmission and conversion, and then air quality is predicted. The quality of the input files of the air quality numerical model affects the accuracy of the air quality prediction. In order to improve the accuracy of air quality prediction, this study proposes a method based on Elman neural network. This method uses Elman neural network to optimize the prediction results of two air quality numerical models of CMAQ and CAMx. First, this study runs the air quality mode CMAQ and CAMx to get the prediction results, and then pre-process the prediction results. The processed prediction data and the measured data are used as the input of the Elman neural network for model training and finally get the neural network model. Through the verification and analysis of the test data set, the experimental results show that the method shows higher accuracy than the single air quality numerical model.

Abstract:As a typical example of China’s invasion of alien species, Eupatorium adenophorum causes serious damage to the ecological environment and affects the development of agro-forestry economy. Eupatorium adenophorum detection as the initial stage and monitoring stage of prevention and control, its detection accuracy will affect the control results. Aiming at the target detection problem of the complex background leaf image of Eupatorium adenophorum, this study proposes a migration learning method based on YOLOv3 to detect Eupatorium adenophorum. The deep learning model YOLOv3 was migrated to the E. adenophorum data set, and the K-means algorithm was used to perform dimensional clustering to determine the target frame parameters. The weight of various losses is changed in the loss function during training, and the adaptability of the model is increased to the data set. The experimental results show that Average Precision (AP) is 17% higher than that of the original YOLOv3 in the detection task of Eupatorium adenophorum, which can meet the detection task of Eupatorium adenophorum under complex background.