This PDF file contains the front matter associated with SPIE Proceedings Volume 12640, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Transmission, transformation and distribution equipment is the foundation of the power network and occupies a very important position in the power enterprise. This paper mainly expounds the application of Internet of Things technology in transmission lines, including the research and design process of transmission line operation environment monitoring technology, which plays an important role in solving the problems of long transmission lines, wide distribution and difficult patrol management. However, there are still some problems in the process of practical application, which need to be paid great attention to. Based on this, this paper discusses the optimized application of information transmission line inspection management system, hoping to provide reference for related research.

In order to solve the problem of lack of reference to real ship trajectory in ship route planning, a method of ship multi-constraint route planning based on AIS (Automatic Identification System) big data is hereby proposed after analyzing existing route planning methods. The AIS feature trajectories generated in the preprocessing stage are clustered using Fuzzy Adaptive Density-Based Spatial Clustering of Application with Noise (FA-DBSCAN) to identify similar ships turning area. Under the constructed navigable grid chart environment model and multi-condition constraints, the ship route planning is realized by improving the ant colony algorithm. The simulation test results show that the planned route obtained using this method is provided with more advantages than other methods in terms of navigation cost, applicability and operating efficiency.

With the rapid development of global technology and the expansion of traditional industries, energy shortages and environmental pollution have become issues that cannot be ignored. Ubiquitous lighting facilities, as a significant feature of modern construction, consume a lot of energy, and with the acceleration of urbanization, the construction scale of road lighting will continue to expand, and energy consumption will also be further expanded. As a key technology for the upgrading of traditional industries, the Internet of Things technology is also a key technology for building a smart city. The unified control and deployment of lighting facilities through the Internet of Things technology can promote the development of lighting facilities in the direction of intelligence and energy saving. This paper first briefly introduces the architecture design of a smart light pole system, and briefly describes the realized functions and an application case in the expressway service area, providing a systematic solution for the construction of smart lighting facilities.

With the gradual formation of a national digital sharing economy and the growing international concern about environmental issues in developing countries, the emergence of a new "smart environmental protection" ecosystem is imperative. In this paper, the future development needs, architecture, and applications of IoT in environmental protection are presented to integrate IoT into environmental protection with the help of computers and cloud computing to achieve the integration of human society and environmental business systems for environmental management and decision-making in a more granular and dynamic manner. The application of IoT + Big Data in smart environmental protection will greatly accelerate the construction of industrial ecology.

With the development of the Internet of Things technology, smart agriculture has attracted wide attention. The IoT project based on the multi-source data fusion algorithm and the Harmony system provides a more convenient and fast way for the actual production of agriculture. By detecting the multi-source environmental information in the production, such as carbon dioxide concentration, light intensity, air humidity, etc., the data is processed by using the multi-source data fusion algorithm. Through Huawei Harmony OS we can build a smart agriculture app and can through the BearPi-HM Nano connection sensor and preliminary prove the feasibility of comparison in this direction of innovation, this can timely and reliably guide agricultural practitioners, reduce the waste of manpower and material resources in agricultural production.

With the advent of the Internet of Things technology, the smart home has been redefined by the concept of "digital home". Especially under the circumstances of global epidemic, the Internet of Things technology can monitor the temperature and humidity of the living environment at any time through sensors, which allows people to know the temperature and humidity in their homes in time, so as to predict the potential danger in advance. The system, The Internet of Things temperature and humidity detection, is mainly designed to collect the data of temperature and humidity in the home and access to achieve communication with Mobile phone. The system, by using the Micro Python language, collects the data of temperature and humidity through the DHT11 temperature and humidity sensor, and achieve communication function with upper monitor with the help of WIFI.

As an industrial field bus, CAN bus has the characteristics of high real-time and high reliability. This chapter analyzes the risk situation by using CAN bus through the research on the actual situation of the industrial field. First of all, we understand and analyze the actual situation of industrial field and CAN bus; Secondly, the risk assessment model is designed and applied to the actual risk assessment process; Finally, the risk assessment model is used to analyze the countermeasures and specific suggestions that need to be improved at present.

The scenic spot has a certain tourist carrying capacity, controlling the number of tourists in the scenic spot can not only protect the scenic spot landscape, but also can improve the comfort of tourists. With the development of modern information technology, the demand for building smart scenic spots is becoming stronger and stronger. Tourist volume monitoring is an important aspect of smart scenic spot construction[1]. The system adopts intelligent camera as the tourist volume collection equipment, which is installed at the entrance and exit of the scenic spot to make real-time statistics of the incoming and outgoing data[2]. Our innovations include the followings. First, we have designed a general system architecture based on intelligent tourist volume camera for tourist volume collection, which is applicable to the statistics of tourist volume of scenic spots at multiple entrances and exits, and supports monthly reports, daily reports and hourly reports. Second, we have proposed a three-level threshold tourist volume early warning mode, which is based on 80%, 90% and 100% of the tourist carrying capacity of the scenic spot, and set three-level threshold, two-level threshold and primary threshold respectively. This is an application of IoT in intelligent scenic spots.

The scale of school construction is becoming larger and larger, which requires that the traditional information management and processing need to be changed to adapt to the continuous development of society. How to use the means different from the traditional way to make corresponding changes and changes, and establish systematic and convenient information management has become an indispensable requirement. Only by establishing an intelligent management system, can data information be carried out effectively and quickly. This not only reduces the workload but also avoids the occurrence of errors affecting the normal teaching work. In this paper, part of the cloud platform data is obtained through the My SQL database provided. According to the cloud platform database, the information is obtained to help analyze the goal of user behavior. The remaining cloud platform data is obtained through the provided cloud platform API interface, and the data processing is collated and executed for subsequent use. This paper designs the management system of distance education for agricultural talents, and initially constructs a new operation model of the design education platform which adapts to the era of mobile Internet and conforms to educational reform. The experiment tells that the related design of this paper plays a positive role in the reform of design education and teaching in practice.

This paper proposes a Mahalanobis-Taguchi system variable optimization method based on binary quantum behavior particle swarm. Firstly, the Gram-Schmidt orthogonalization method is used to calculate the Mahalanobis distance value. Through the ROC curve, the optimal threshold point of the system classification is determined. The misclassification rate and the selected variables rate are defined, the multi-objective mixed 0-1 planning model is built.The improved quantum behavior particle swarm optimization algorithm is proposed to solve the optimization combination. To adapt to the binarization variable optimization problem, the algorithm performs binary coding on the particle based on probability. Using the optimized combination of variables, a new Mahalanobis-Taguchi metric basedprediction system is established to complete the task of precise discrimination.

Water body extraction is a significant direction of application in information monitoring of water resources. Thus, on the basis of high-resolution UAV remote sensing images and self-build inland water body datasets, this paper designs a water body extraction method fused atrous spatial pyramid pooling. First, SegFormer detection network is improved, and the semantic segmentation network is adopted to mine water body semantic information of high-resolution UAV remote sensing images; second, the atrous spatial pyramid pooling decoding module is fused to explore multi-scale contextual information, solve the problems such as feature information loss of water body and deficiency of long-distance information, and come true the coverage of more geometric information whilst responding to semantic features; finally, the semantic segmentation dataset of the inland water body is self-established on the basis of the UAV remote sensing images in Dengzhou, Henan, and the experiment is conducted. Then the validity of the proposed water body extraction method is validated via the comparative ablation experiment with Hrnet, PSPNet, UNet, Deeplabv3+ and the original detection algorithm. According to experimental results, the improved method is superior to the comparative detection methods such as UNet, and has a better effect on the integrity detection of water body edges and the reduction of missing detection and false detection of small water bodies and tributaries.

Aiming at the lack of intelligent comprehensive monitoring and data analysis of environmental meteorology in agricultural production activities, a monitoring platform integrating IoT technology is designed for smart agricultural meteorological station. A three-layer system of sensor terminal, IoT convergence gateway and cloud monitoring platform is constructed to realize real-time uploading and recording of agricultural meteorological data through Internet wireless communication module. A cloud platform and visual monitoring background which can support the concurrent access of more than 1000 distributed meteorological monitoring stations are designed and developed. The local maintenance of the weather monitoring station is realized by the mobile APP, which can set and modify the parameters of the station. Through practical application tests, the environmental meteorological parameters can be monitored in real-time, comprehensively and accurately, and the meteorological data can be recorded flexibly by the platform. The platform is convenient to operate and maintain, and its functions are easy to expand. It can provide important data support for the development of agricultural activities and play an important role in promoting the mature development of smart agriculture.

With the rapid development of the science and technology service industry, the number of modern science and technology resources is increasing. In the current science and technology resource recommendation process, the content-based recommendation method is mainly applied for resource filtering. The resource scoring results are easily affected by data sparsity so that the F-measure value of the personalized recommendation results of the system is low. Therefore, this paper proposes a personalized recommendation system of modern science and technology resources based on the hybrid filtering algorithm. Starting from the user's basic information, browsing time, and other data, the multi-dimensional user characteristics are obtained. A bipartite network containing several users and resource items is constructed, and a user-resource rating matrix is defined. Based on the concept of hybrid filtering, the dynamic weighted calculation is carried out on the predictive scoring results of collaborative filtering and content filtering to obtain a reliable comprehensive scoring prediction result, based on which all modern scientific and technological resources are filtered. The recommended resources are sorted in descending order according to the scores of multiple users. The top resources are selected as the personalized recommendation results. The system test results verify that the F-measure value of the proposed system is 0. 85, which meets the requirements of personalized recommendation of science and technology resources.

With the development of the power grid, the types and number of power intelligent terminals continue to increase, which brings great challenges to the operation and maintenance management of power grid IOT equipment. In order to help users, locate and solve problems encountered in the terminal access process quickly, we design a troubleshooting tool based on the structure of the IOT management platform. Through this tool, we can simulate the actual situation, to help users locate the fault location and find solutions efficiently and accurately. From the practical application effect, the tool can shorten the time for troubleshooting and significantly improve the work efficiency.

The power industry has achieved rapid development with the strong support of national policies, which makes the relevant data information show geometric growth. With the further promotion of the power market reform, the market transaction subjects are gradually diversified. In this progress, a good trading strategy is particularly important in order to get more generous profits in the market trading progress. The application of power big data technology can provide security guarantee for electric power production and electric power transaction. Through data mining, data analysis, data extraction and data storage, it can provide a data basis for the development of various transactions in the electric power market. Based on the mining and processing of power big data, this paper builds a two-stage decision-making framework for the power market to study the trading strategies of market participants. The model established in this paper considers the uncertainty in the two-stage decision-making process, which can reflect the risks associated with the decision-making process.

With the progress of science and technology and the development of economy, especially the rapid development of modern logistics industry, the wide use of various rich logistics digital resource database, we have the experience of big data era has come; many artificial operation process has failed to meet the needs of the development of The Times, can not meet the growing demands of diversified logistics information management. Once we enter a query data in the process of selecting logistics information, the result is either a mass of query data, or elusive useless data, really useful data query is very difficult to oneself, it is difficult to find accurate logistics information; how to find suitable and useful data in the face of massive big data, is the key to the current research problem.

With the deepening and acceleration of urbanization and motorization in China, the problem of urban traffic networks represented by traffic congestion has become increasingly prominent. As an important part of the urban rail transit system, the metro plays a significant role in alleviating the pressure on road traffic. Accurate metro traffic flow forecasting can provide important guidance for the optimization of urban transportation network layout. In this paper, a novel traffic system coupling singular spectrum analysis, deep bidirectional LSTM networks, and ensemble strategy is established for metro traffic flow forecasting. Real metro traffic flow data from three different areas of Shanghai are used to verify the validity of the model. The experimental results show that the proposed coupling system can accurately predict the changes of metro traffic flow.

The emergence of O2O e-commerce model has put forward higher requirements on the cold chain logistics distribution capability of fresh produce e-commerce. This paper constructs a fresh food logistics optimization model under the O2O e-commerce model,uses a genetic algorithm to solve the "last mile" optimal path, and MATLAB software is used to verify the effectiveness of the algorithm applied to a the optimization of the end delivery path of a fresh food shop. Finally, the research method and the fresh produce e-commerce operation are suggested to provide reference for the fresh produce platform to achieve "cost reduction and efficiency enhancement".

The mode of educational reform and development has undergone profound changes. To turn teaching into a knowledge-intensive specialty, it is necessary to construct an ecological environment that is conducive to the creation, dissemination, sharing, and accumulation of educational resources. This will be conducive to promoting the sharing and flow of educational resources within the organization, as well as the updating of knowledge and the professional development of individual teachers. At the same time, it can also promote the positive transformation of teachers' individual knowledge to organizational knowledge, and promote the systematic construction and optimal development of the organization. This paper integrates information technology and primary educational resources to promote the sharing of high-quality resources to solve the problem of uneven distribution of primary education resources. Feature clustering is integrated into multi-task recommendation, and neighbor subsets are recommended to share real information according to the quality index of nodes. This strategy provides external incentives to the nodes to help them evaluate the trustworthiness of the nodes' information sharing behaviors in their neighbors. In this paper, it is recommended that a subset of nodes with high quality index share information, which plays a role in encouraging efficient information cooperation among nodes. The results tell that the proposed clustering algorithm and the multi-task clustering recommendation algorithm have excellent performance in the network. The proposed algorithm has good robustness and accuracy both in the difference of the parameters to be estimated and in the multi-task environment. The research content of this paper is conducive to the high-quality sharing of primary education resources, and plays a role in promoting the improvement of education level.

With the continuous development of China's economy and society and the advancement of the rural revitalization strategy, the agricultural economy is thriving. As an important component of the national economy of the agricultural economy, whether agricultural products can ' go out ' is one of the most important issues. Especially affected by the current new coronavirus epidemic, unsalable agricultural products have been common. Therefore, promoting agricultural products to better ' go out ' has become the primary solution, which shows the importance of logistics and transportation for rural economic development. Based on 8 indexes such as gross agricultural output value, total length of postal routes, and rural delivery routes in Anhui Province from 2011 to 2020, the grey correlation analysis method was used to study the correlation degree between them and the turnover of goods in Anhui Province. The results are as follows: rural delivery routes (0.777) > road operating vehicle ownership (0.765) > gross agricultural product (0.711) > road cargo operating mileage (0.704) > transportation, warehousing, and postal industry (0.7) > civil cargo vehicle ownership (0.697) > railway cargo operating mileage (0.688) > total length of postal routes (0.52). Among them, the correlation between rural delivery routes and cargo turnover is the highest, and the total length of postal routes is the lowest. At the same time, this paper combines the grey prediction model to estimate the cargo turnover of Anhui Province in the next five years. Finally, the paper gives relevant suggestions based on the above data results, which have certain practical reference significance for improving the existing logistics and transportation problems of agricultural products in Anhui Province.

With the rapid development of China's economy, the total amount of energy consumption in China is also increasing. It is a major task to achieve energy conservation and green development. This paper is based on the historical data of China's energy consumption structure from 2007 to 2020. First, based on Markov chain theory, a prediction model of energy consumption structure is constructed; Secondly, the average transfer probability matrix of the energy consumption structure is calculated, and the optimal transfer probability matrix is selected according to the optimization idea; Finally, the optimal transfer probability matrix is used to predict China's energy consumption structure from 2021- 2025. The predicted changes in energy consumption structure can provide relevant reference for optimizing the energy consumption structure.

Aiming at the characteristics of complex business scenarios and sensitive data of distribution IoT, this project proposes a security risk assessment method for distribution IoT based on the analysis hierarchy process, which assesses the security of distribution IoT in terms of hardware, system, application and data. First, according to the characteristics and business scenarios of distribution IoT, the security status is divided into two parts: self-security and security control to ensure the integrity of the analysis. Based on the analysis results, typical security threats of different types of devices are extracted; secondly, based on the security threat analysis results, a security index system is designed and proposed from the perspective of defense and response of distribution IoT, and the index system is classified and quantified to form numerical information that can support the security assessment; thirdly, based on the analysis hierarchy process and fuzzy comprehensive analysis technology, the security risk assessment method of distribution IoT is formed Finally, the effectiveness of the proposed security assessment method is demonstrated by selecting the actual application scenarios of the distribution IoT.

With the development of information technology, the teaching of IoT technology has gradually become popular, but as an important part of hands-on training, the practical aspects of the course are restricted by various factors such as equipment and venue, resulting in unsatisfactory results. Through the introduction of virtual simulation technology in teaching, the IoT programming course is no longer constrained by the above factors, so that the course can reflect the characteristics of IoT applications, improve the implementation effect and teaching quality of the IoT programming course, and stimulate students' interest in learning.

If the Probabilistic Neural Networks (PNN) based on the classification method of equal size interval in training data is used to diagnose the break size of the marine nuclear power plant, the accuracy of the diagnosis results is low when the break size is small. Therefore, a diagnosis method for the break size of a marine nuclear power plant based on the classification method of variable size interval and the PNN is proposed. First, the break sizes are classified according to the variable size interval. Then the data under different break sizes is generated, and the PNN is used to learn it. Next, the corresponding operation data is generated as the real-time data. Finally, the PNN model is used to diagnose the break size. The above processes are repeated with three different variable size interval classification methods, including the break size increasing proportionally, the break size interval increasing proportionally, and the break size interval increasing by arithmetic progression. The diagnosis results are compared with the classification method of equal size interval. And finally, the different classification methods of break size are combined for analysis. The results show that the use of variable size interval classification method that the break size interval changing according to arithmetic progression can increase the accuracy of diagnosis results by 1.21%, and combining it with the classification method of equal size interval can significantly increase the accuracy by 7.21%.

The models used in most deep neural networks SAR object identification algorithms are often designed based on coarse-grained recognition tasks and are not well adapted to the task requirements of SAR vehicle target recognition. To further enhance the accuracy of identification, this article introduces the fine-grained recognition idea into SAR target recognition and designs a bilinear feature fusion convolution module with attention allocation capability. The deep features are first extracted using the residual network, then the attention module is connected to screen the channel information. Finally, the information fusion of the two streams is achieved, and the fused features are used to complete the recognition task. The model is validated on the public dataset. Comparative experiments show our innovative algorithm has a great performance improvement over other classical target recognition methods. The visualization features show that the features mined by this model has better interpretable.

As the core technologies represented by Internet of Things (IoT) technology and computer vision technology booming, object detection algorithms based on deep learning have received high attention in various fields. Object detection is not only a core problem in the field of computer vision, but also a prerequisite and basis for many computer vision tasks, and has important applications in the fields of autonomous driving and video surveillance. The task of object detection is to find all the targets of interest in an image and determine their category and location. Object detection includes two tasks: object classification and object localization. However, object classification focuses on the most discriminative part of the feature map, and object localization requires a feature map focusing on the whole region of the object. Therefore, in order to improve the detection accuracy, we propose a CBAM-RetinaNet network model.CBAM (Convolutional Block Attention Module) is a lightweight convolutional attention module, which combines channel and spatial attention mechanism modules. The channel attention module focuses on the meaningful information in the input image, which would lead to an increase in the accuracy of object classification. The spatial attention module focuses on the location information of the object, which may improve the accuracy of object localization. The experiment results show that the proposed achieves better object detection performance on the PASCAL VOC2007 dataset, and the accuracy is 2% higher than that of the RetinaNet model, which has good experimental results.

Internet of Things (IoT) and AI are correlating with each other closer and closer. By combining these two techniques, most of the tasks can be done efficiently. In this circumstance, however, most of the recognition tasks of handwritten Tibetan is still fulfilled manually or using methods like convolutional neural network (CNN), which lowers the efficiency because of their long training time. In order to solve those problems, in this paper, we use portable IoT devices and seek for a traditional method with not only high accuracy but also short training time on the tasks. We use methods that are commonly used in handwriting recognition to recognize handwritten Tibetan numerals and compare not only accuracy but also training time of those methods. Furthermore, we adjust the best method and find the most appropriate setting for handwriting recognition task. We combine AI and IoT and provide an efficient way of recognizing handwritten Tibetan numerals.

To solve the problem of low classification accuracy of electricity safety status on the customer side, which affects the monitoring accuracy, this paper studies the monitoring method of electricity safety status on the customer side combined with the perception technology of the Internet of Things. The power consumption monitoring data on the user side is collected by using the IoT perception. The TAN network topology is taken as the core to construct a security state identification model, and the model operation result is output as the monitoring result to realize the power consumption security state monitoring. Through experimental verification, the new monitoring method has higher accuracy in safety state classification and higher monitoring accuracy, which can comprehensively improve the perception, monitoring and service capabilities of customer-side electricity safety, and enhance the level of user electricity safety.

To solve the problems of low detection accuracy, large amount of parameters and computation, a fast pig detection algorithm based on yolov5s is proposed in this paper. First, to enhance the learning ability of feature extraction, RepVGG Block is used to replace the ordinary convolution in the backbone network, and proposes a new feature extraction structure named R-CSPD that combined CSPDenseNet and RepVGG Block. Second, the detection speed is increased by compressing the number of channels in the neck network and using depthwise separable convolution to reduce the number of parameters and computations. Finally, C3VGG is used to optimize the C3 structure to enhance the localization ability of the target. The experimental results show that compared with the benchmark model, the calculation amount of the model in this paper is reduced by 30.7%, mAP@0.5 is increased by 1 percentage point, the mAP@0.5:0.95 is increased by 2.4 percentage points, and the detection speed is increased by 17 FPS.

With the development of wireless network technology, it becomes possible to transfer huge amount of data from the center to the edge server computing. When deciding what kind of data tasks to offload to the edge server, we need to make a prudent decision. Traditional mathematical computing methods are computationally intensive, require computation all the time, and have high processor hardware requirements. We use the Actor-Critic algorithm to model the edge computing environment as a binary offloading process. Applying machine learning algorithms to the edge computing environment allows the intelligence of machines to replace a large number of mathematical operations. The Actor-Critic algorithm is highly adaptive and can effectively solve the current problems faced by edge computing, such as large computational volume and narrow application surface.

Based on the existing time domain convolution module, attention mechanism, relationship network and other artificial intelligence technologies, the existing virtual reality products are aimed at the problems such as picture dragging and eye fatigue caused by low refresh rate, and the existing virtual reality products do not set sleep mode for people with sleep disorders. This paper proposes an adaptive black insertion display technology based on human eye activity. Users can switch game mode and sleep mode according to their own needs.

With China's entry into an aging society, the number of the old is increasing. Due to their weakened physique, the old often need to take a variety of drugs without special personnel to take care of them, and such phenomena as missing or overdose often occur. This paper designs an intelligent medicine box based on MCU and WiFi technology, which is remotely controlled by a mobile phone. The system takes the STC12C5A60S2 chip as the control core, sets the medication information through the mobile phone APP or key circuit, uses the clock chip to time and display information on the LCD screen, and realizes the functions of the medicine box timing pop-up, voice broadcast reminder, human-computer interaction, etc. This greatly reduces the occurrence of patients forgetting to take medicine, taking medicine repeatedly and taking medicine wrongly, etc.

In multi-round dialogue systems, we can easily find that the final reply is closely related to two points, one is the context of the dialogue, the other is the persona characteristics. But not all characters and contextual information will affect the final reply, because the final reply may only be related to some crucial characters and contextual information,the indiscriminate use of all information may even have a negative impact on the generated dialogue. So it is necessary to extract and utilize key characters and contextual information to improve the quality of the final generated response. In this paper, we show how to solve this problem through our new model and methods. Specifically, our new model consists of two parts: encoder and decoder. The encoder is mainly used to encode personas, contexts and historical responses, and the decoder generates corresponding words from the vocabulary. Then, the weight of the character and context is updated through the multi-head self-attention mechanism to affect the response generated by the decoder. The experimental results show that compared with the baseline models, our model and methods have improved in terms of metric-based evaluation.

Malware is still a big threat to network security, and various types of malware detectors are needed. Deep learning-based classifiers have substantially improved their ability to identify malware samples. However, these detectors suffer from adversarial examples. The samples were made by adding small, carefully selected perturbations to the normal software. Any vulnerability in malware detectors can pose a significant threat to the platforms they defend. However, existing attack methods may not meet the inherent limitations of malware. This paper proposes a new method to generate malware adversarial samples. The original malware is mutated into new samples through semantic analysis of malware, transplantation of code in the program and addition of code at the end. And used to fool the detector. Experiments show that compared with the existing methods, our method has a significant effect on the efficiency of generating adversarial examples and the success rate of the attack.

In this paper, we propose the YOLOv4-Sensitive algorithm based on the YOLOv4 algorithm. Firstly, the residual cell structure in the backbone feature extraction network CSPDarknet53 is reconstructed, and a new feature extraction network U-CSPDarknet53 is designed to extract and retain small target feature information at finer granularity. Secondly, MFENet (Multi-receptive Field Extraction Network) network is designed to extract the contextual information of small targets using parallel expansion convolutional branches to alleviate the feature loss problem of SPP network due to pooling operation. Finally, CA attention mechanism is introduced to design a multi-scale feature fusion network CA-PANet to integrate location information into the feature aggregation network and enhance the feature description capability. The algorithm is validated on the PASCAL VOC dataset. the improvement of YOLOv4-Sensitive for small target detection accuracy is more obvious, with an improvement of about 4.02 percentage points. In addition, the inference speed of the algorithm in this paper is 42 frames/second, which improves the small target detection accuracy without losing the inference speed.

With the proposal of new power system, the construction of the power Internet of Things is also accelerating gradually. As an important equipment for collecting massive data, how to ensure the information security of the Internet of Things perception layer terminal has become an urgent problem to be solved. We proposed a secure access scheme for electric Internet of Things terminals, which includes three modules: secure SDK, terminal encryption management module and agent component. The secure SDK is used to realize terminal security encryption and decryption. The terminal encryption management module is mainly responsible for file and information management, and the agent component is mainly responsible for device configuration update. Based on this scheme, the terminal accesses the secure access gateway by changing device configuration information through the agent component, after the authentication of the unified password service platform is completed, the terminal encryption management module sends the digital certificate to the terminal remotely, the terminal establishes an encrypted transmission channel with secure access gateway through secure SDK to ensure the security of data transmission.

To determine the relationship between various factors leading to ship collision accidents and prevent ship collision accidents. This paper collected 100 reports of ship collision accidents, and 40 causative factors were determined from the four aspects of human, ship, environment, and management. The relationship among 136 kinds of causation factors was determined by extracting the causation chain from the accident report, and the causation network model of ship collision accidents was established. Python simulation was used to analyze the robustness of the causative network under deliberate and random attacks. By comparing the robustness of the causative network under degree value attack, betweenness centrality value attack, closeness centrality value attack, and PR value attack, the key factors in the causative network were identified, and corresponding prevention strategies were proposed. The simulation results show that the robustness of the causative network is weak under the deliberate attack, the robustness of the betweenness centrality value attack is the worst, and the node with a higher betweenness centrality value is the key node in the causal network, giving priority to the prevention of key nodes is conducive to the safe navigation of ships.

The new power system can strengthen the resource allocation ability of the distributed Internet of Things, and can also make the intelligent coordinated control of a large number of decentralized power generation and supply objects. It is an urgent task to realize the two-way interaction of various market players and establish a corresponding power communication network for the realization of a new power system. Aiming at the quality problem of communication data transmission of power Internet of Things (IOT) equipment, this paper proposes a unified data model of IOT agent. It can encapsulate the collected data into a user-defined binary data packet format for transparent transmission. This avoids complex data parsing and format conversion and improves the efficiency of data transmission. To better reflect the feasibility and effectiveness of the IOT agent information bearing model, this paper applies different network scenarios for simulation analysis. The experimental results indicate that the model can reduce the data transmission delay and the packet loss rate.

foreign matter intrusion on train operation safety is becoming increasingly serious. Based on a large amount of research data, among many factors affecting the operation safety of high-speed railway, the detection of personnel and foreign matter intrusion is difficult, and the preventive measures are limited, mainly because of its strong randomness and complexity. This paper systematically introduces the development status of high-speed railway line protection technology at home and abroad, especially summarizes the development status of high-speed railway line monitoring technology in China, and proposes a new design of foreign object intrusion detection and alarm device based on new sensors and networks, which has certain reference value for high-speed railway line protection and railway bureaus and departments in related fields.

The spatial and temporal characteristics of regional traffic flows and their socio-economic significance are investigated by using machine learning and ArcGIS technology with multiple attributes of highway toll station data. The study found that: (1) the overall characteristics of the traffic flow at 10~11, 14~16 and 17~19 hours show a "three-peak" structure, while the spatial distribution of high, medium and low traffic types has obvious clustering characteristics. (2) The specific features of the K-means++ based highway toll station classification are "point-line surface" structure in space, with Kunming West Toll Station, Dujiaying Toll Station, Kunming North Toll Station and Liangmiansi Toll Station as unique Node, Kunchu line and other toll stations along the axis, the rest of the toll stations constitute the surface; time, each type of toll station traffic flow also shows the "three peaks" structure, but there are "peak" and "sub-peak The "peak" and "sub-peak" are divided. (3) Based on ArcGIS technology, the dynamic visualization spatial expression of traffic flow "three peaks" separated by time series reflects the distinctive "day-night" pattern of human travel activities across regions.

At present, many manufacturers in China blindly implement reverse logistics without proper planning, which leads to a large number of waste household appliances flowing into informal recycling channels and dismantling markets, causing resource waste and environmental pollution. Therefore, this paper studies the construction of reverse logistics network of waste household appliances, establishes a mixed integer programming model with the goal of maximizing revenue, and verifies the effectiveness and rationality of the model and algorithm by combining genetic algorithm and taking Company M as an example.

Container yard is a key link for terminals to achieve carbon neutral development, and yard crane is an important equipment used in the yard. Taking a single yard crane as an example, the optimization model of yard cranes scheduling in container yard is established by considering the cost of moving the yard crane, the number of transitions, the waiting time of container trucks, and the carbon emission generated by the yard crane during moving, loading and unloading as well as waiting, and the model is solved by using non-dominated sorting genetic algorithm (NSGAII), and finally the validity of the model is verified by using example data. It is expected that the research in this paper will be of positive reference significance for improving the efficiency of yard crane operation, reducing the carbon emission from yard crane, and realizing the carbon neutral development of the port.

In this paper, one kind of polarization independent terahertz perfect absorber of narrow band based on cross and open ring structure of graphene is proposed. The effects of Fermi energy level and structure size of graphene on the absorption characteristics of meta-material absorber were studied by using CST electromagnetic simulation software. The physical mechanism is explained by impedance matching theory. The results show that when the Fermi energy level of graphene is 0.5 eV, the radius of inner ring is 8 μm, the radius of outer ring is 11.5 μm, the width of cross is 1.5 μm, and the thickness of silica dielectric layer is 10 μm,, the absorber can realize 98% perfect absorption in 4.48 THz. Moreover, due to the symmetry of the structure, the absorber is independent of polarization. This result has great applications for the realization of ultra-thin tunable smart devices in the terahertz band by using patterned graphene.

Conveyor is one of the main equipment for coal production and transportation. Due to impact fatigue, uneven surface stress of conveyor belt and other external factors, deviation will occur, leading to material overflow and damage to transportation equipment. Therefore, it is of great significance to detect the deviation status of the conveyor belt quickly and timely to ensure the safe and efficient operation of the transportation system. This paper presents an automatic detection method of belt deviation based on DeeplabV3+, which can detect the deviation of any position of belt conveyor. We have established a new belt edge dataset under real working conditions. In order to improve the deviation detection accuracy, we expand and erode the image after feature extraction, extract the centerline, and finally detect the deviation distance through the deviation detection module. Experiments show that this method can well balance the detection accuracy and detection speed. The processing speed of a single image is 0.32 s, and the conveyor belt edge detection error is less than 6mm, this method has good real-time performance and high precision, and can be applied to the production scene of underground coal mine.

In the aerospace field, in the process of identifying and tracking air targets, it is faced with inaccurate or incomplete target recognition. This research is based on the auxiliary analysis and recognition capabilities of deep learning convolutional neural networks to realize the recognition of aerial targets, so as to improve the recognition accuracy and reduce the recognition error rate. Take the aircraft data provided by the Institute of Aeronautics and Astronautics as the research object, perform image preprocessing on it, build the aircraft data set, build a network framework using python language in the TensorFlow environment, and perform recognition training on the model, and finally test the trained model and result analysis.

In practice, the effluent quality is the most important indicator to measure the performance of wastewater treatment. However, most existing sensing technology are difficult to maintain the real-time accurate of online monitoring. In his paper, a novel no-inflection point neural network (NI-BP) with adaptive learning rate is proposed for wastewater quality indicators prediction. The proposed algorithm updates the weight parameters by an adaptive learning rate method when some inflection points are detected, it can increase the weight iteration step size and make the network converge quickly. In the simulation examples, the data from Benchmark simulation model NO.1 (BSM1) are used to demonstrate the effectiveness of the proposed method. The results show that the proposed NI-BP can significantly accelerate the model training efficiency and the prediction accuracy, simultaneously.

The Internet of Things is the product of the information technology revolution and has a broad application prospect in all walks of life. The intelligent hydroponic greenhouse is not only the product of scientific and technological development, but also the development trend of future agriculture. The intelligent hydroponic greenhouse using Internet of Things technology can not only realize automatic monitoring, intelligent control, remote control and other functions, but also make the development of agriculture more intelligent and modern. Two monitoring networks of nutrient solution and greenhouse environment are respectively built through the Internet of Things to realize real-time monitoring and control of nutrient solution and greenhouse internal environment, truly liberate productivity, greatly reduce the labor intensity of users, not only make agricultural development more accurate and intelligent, but also make agriculture more integrated and industrialized.

Waste recycling industry is a promising industry in China. With the rapid increase of China's population, a large number of products are produced, resulting in more and more waste products with more and more complex types, many of which can be recycled. This design adopts STM32 MCU based on ARM architecture to realize the design of intelligent waste recycling station. This design is not only in line with the social development trend, but also the inevitable trend of sustainable development. Compared with the traditional way, it not only saves time and improves efficiency, but also conforms to the development trend of smart life, which is conducive to improving people's living standards, improving the construction of residents' infrastructure and making residents' lives more intelligent.

The surface viscoelastic model was obtained by cell surface mechanics analysis. On this basis, the mechanical data of hepatocytes were obtained by nanoindentation experiment. BP neural network was used to train the mechanical characteristics of cell surface, and the mechanical relationship of stress and strain was obtained. Aiming at the common problem of gradient descent trap and convergence rate of BP neural network, genetic simulated annealing algorithm is introduced to optimize the network weight and improve the prediction accuracy of BP neural network. Finally, the error rate of the neural network model and the traditional model was compared by full scale fitting, which proved that the mechanical model optimized by the neural network could describe the mechanical properties of the cell surface more accurately.

This research suggests a fast lane line identification approach based on deep learning to address the issue of vehicle occlusion in lane line recognition as well as the poor speed of lane line detection. The backbone network's feature extraction speed is increased with the usage of a deep separable convolution technique. In order to address the issue of insufficient lane line recognition when the vehicle is obstructed, the Feature Pyramid Networks (FPN) approach is utilized to improve the extraction of contextual information from the network. To accomplish quick and precise lane line recognition, the lane line thin structure characteristic is completely leveraged, the lane line a priori approach is applied, and the line IoU (L-IoU) idea is used to introduce line IoU loss. The accuracy rate on the Tusimple dataset achieves 0.9652 on the CULane dataset 76.11 F1 score, and the detection speed is 212.5 Fps.

Cardiovascular disease has dominated fatality cause in recent decades, the most vulnerable group of which, is reported to be middle-aged and older men based on much clinical research. This article intends to investigate the underlying reasons for such a group's susceptibility through the combination of empirical evidence and data analysis by logistic regression. In conclusion, it turns out that unhealthy habits such as chronic smoking, unlimited drinking, and the pronounced lack of estrogen protection exacerbated by the aging effect are the crucial reasons why there are more middle-aged and older male patients than females. In the final section, some tailored precautionary measures have been provided functioning as the relief of the aging-heart burden.

At present, it is difficult to distinguish the relationship between the mechanical properties and process parameters of composites by direct calculation or finite element method. But, through the neural network method, there is no need to involve the solid-liquid coupling problem. In this paper, via the study of neural network, the model of relationship between the process parameters and the mechanical properties of the composites is established. The calculated results of the model are in good agreement with the experimental values after training the classifiers by the samples, and its show that the model is correct. At the same time, a set of data is used to test the model, and the results are consistent. This study also shows that the interlaminar shear strength and bending strength of the composites increase with the decrease of resin injection temperature，shortening of resin injection time and the increase of resin injection pressure.

After implementing the Jaminan Kesehatan Nasional (JKN) in Indonesia, health system inequity, payment non-compliance and additional expenditure still exists. To better deal with the problems in their healthcare system, this study uses a variety of machine learning algorithms to classify patient blood samples for improving the efficiency of healthcare system. The study shows that most of the algorithms are up to 70% accuracy and the accuracy will rise with only important variables.

Handwritten character recognition belongs to the image feature classification problem. To address the problems that convolutional neural network-based handwritten character feature extraction algorithms have complex structures, long training time and recognition accuracy of similar characters is much lower than the average recognition accuracy, we improve the VGG16 and propose a new character recognition algorithm, which effectively improves the recognition accuracy of similar characters. The improved model reduces the complexity and accelerates the model convergence by simplifying the original convolutional structure, adding Ghost convolution, and batch normalization layer. In addition, we introduce an attention mechanism and deformable convolution to solve the problem of false recognition caused by local feature differences and stroke deformation of similar characters. Experimental results on Chinese character datasets and self-built datasets show that our proposed approach achieves over 97% accuracy in general and 92.82% in similar characters. Compared with the traditional approaches, the training speed is faster, and the similar character recognition effect is improved. The model size is reduced by 80%, achieving lightweight and having important application value for improving the performance of character recognition.

People's daily lives are getting more and more entangled with the Internet as technology advances and the level of living rises. The diversity of mobile Internet applications has also attracted some unscrupulous elements to use counterfeit APPs, fraud-related APPs, gambling-related APPs, and other forms to obtain ill-gotten gains, and users have suffered huge economic losses as a result. To solve the current problems, it is especially important to extract APKs quickly for such APPs and the subsequent related analysis of APKs through relevant tools and determine whether they are involved in fraud. In this paper, we propose a dynamic and static method combining unknown APK download links, running screenshots, and text data to detect whether unknown APKs are involved in fraud and to classify fraudulent APKs in detail.

With the continuous improvement of teaching methods, people put forward higher requirements for the quality of English teaching. Traditional approaches for assessing instructional quality are less sensitive to the initial data in the evaluation process, and cannot achieve the optimal convergence effect. To better evaluate the level of instruction in English, the research uses the improved EM algorithm to create a diagnostic evaluation model for English teaching quality. The performance of the model is tested by using the English scores of college students. The outcomes demonstrate that the accuracy of enhanced model is 99%. The trained model is used for data testing, and the test error obtained is 0.8%. The error is 0.4% and 0.1% lower than EM model and FCM-EM model respectively. Therefore, the IFCM-EM evaluation method for teaching English proposed in the study has good accuracy and efficiency. It can effectively diagnose and evaluate the effectiveness of English instruction in schools, as well as offer assistance with improve English teaching evaluation methods and teaching levels.

Product placement is a key part of the whole tobacco industry chain. Traditional placement often costs a large manual workload and is strongly influenced by personal experience. To achieve accurate cigarette placement, we propose a set of data-driven intelligent strategies in different segmented markets. An accurate retailer classification cigarette placement algorithm based on the attributes of business circles is proposed, business circle data outside the tobacco database is introduced, and a neural network cigarette placement algorithm with the fusion of business circle features is established. At last, the experimental results show that the average accuracy of the brand (Zhenlonglingyun) based on feature fusion is 84.7% and the brand (Nanjingyinghong) accounts for 90.1%. Through data cleaning and feature fusion, the deep learning model can be trained to generate customized marketing strategies and achieve intelligent and accurate cigarette placement.

In this paper, C5F10O (perfluorpentone, C5 for short) is decomposed in GIS/GIL due to the existence of insulation defects. C5F10O will produce typical decomposition gases such as CO, C3F6, CO2, C3F8, C2F6, C2F4 and CF4 under overheating conditions. The concentration of the above gases under the overheating conditions is a strong correlation series of time. It is necessary to carry out high-precision fitting for the time series obtained from the above tests. First, Gauss Chebyshev process is used to reconstruct the background value, which realizes high-precision fitting and prediction and early warning of two-dimensional nonlinear time series. At the same time, the maximum likelihood estimation of three-dimensional curved surface combined prediction model is realized. This paper first analyzes orthogonal prediction model based on the grey Markov chain Gauss Chebyshev from the mathematical model level, then analyzes the decomposition characteristics of C5F10O/CO2 under local overheating conditions and its prediction technology, and obtains the optimal ratio of new environment-friendly gas insulation, which has good guiding significance for the application of the new insulation gas to the operation, the maintenance and repair technology of the high-voltage electrical equipment.

Due to interactions with real users, online reinforcement learning training projects for dialogue agents are expensive. User simulator is an alternative method that is commonly used. However, the environment of a user simulator is not identical to that of a real user, and it cannot provide the atypical and more variegated conversational behavior that is a hallmark of human spontaneity. We employ offline reinforcement learning and Transformer to abstract dialogue policy as a framework for sequence modeling problems, modeling the joint distribution of state, action, and reward sequences to generate optimal dialogue actions. An evaluation of the Multiwoz dataset shows that DT successfully improves the efficiency of DRL dialogue agents and improves dialogue robustness.

Heart disease is a serious threat to human health and devastating to families. Sudden heart attacks are often difficult to treat, so it is important to detect and prevent heart disease early. Most of the previous heart disease prediction methods are based on statistical machine learning algorithms, which can achieve good results but can only learn some superficial representations due to their own features, and cannot capture deep relationships, so to solve this problem we use deep neural networks for heart disease prediction. Further, current neural network methods often have difficulty learning discriminative features related to heart disease prediction. Therefore, to solve this problem, we leverage the center loss to enable the neural network to learn discriminative features and separate samples from different categories. We conducted experiments on one dataset, and the experimental results show that our method can effectively improve heart disease prediction while learning more discriminative features.

Facing complex and diverse action recognition scenarios, artificially designed neural networks show poor generalization performance. Therefore, an automatic designing method of 3D convolutional neural networks based on neural architecture search is proposed. Firstly, a variety of human behaviors are extracted to construct training sets and validation sets. Additionally, the weights of neuron connections are updated using the loss on the training set, the discrete network architecture search space is continuous through continuous relaxation, and the search space is reduced by using the hierarchical idea. What’s more, the objective function is optimized by combining gradient descent, realizing fast search, and stacking the obtained computing units to form an overall network at the same time. Evaluations based on public data sets show that the designed neural network model achieves comparable performance to the artificially designed network model in the task of human action recognition, solving the difficulty of deep learning method migration between working at different scenarios by automatically customizing the network.

Regional cargo transportation optimization is the key to the overall efficiency improvement of logistics. Heavy trucks are especially important in connecting regional long-distance transportation and heavy cargo transportation. Relying on transport flow data, we take 6-axle trucks as an example, and build a K-means clustering model to label freight vehicle groups based on the analysis of truck trip intensity and travel time differences, and subsequently design the Random Forest-Recursive Feature Elimination (RF-RFE) algorithm to rank the importance of freight features, and use the filtered feature indicators to verify the travel differences of different groups.The results show that (1) heavy-duty trucks have a higher proportion of nighttime trips, staggered features with other models, and assume more medium and long-distance transport functions; (2) from the K-means++ clustering results, six-axle truck transport can be divided into three types: heavy-duty long-distance transport type, heavy-duty short-distance transport type and light-duty short-distance transport type. (3) RF-RFE model feature ranking in vehicle weighing and travel distance importance ranking the top two, ranking correct rate higher than up to 91%, indicating that loading and travel distance can effectively distinguish heavy-duty truck operation characteristics.

The prediction of carbon price can contribute to the reduction of carbon emissions. A carbon price forecasting model based on the data decomposition method, adaptive boosting (AdaBoost) algorithm, and Elman neural network (ENN) is proposed in this paper, which firstly decomposes original data into subsequences by the variational mode decomposition algorithm, and then combines the ENN models by the AdaBoost.RT algorithm to forecast each subsequence, and finally the predicted results of each subsequence are combined into the final prediction results. Using the carbon price in Beijing, China as the experimental data, the evaluation errors certify that the proposed ensemble model can achieve a better forecasting effect than the single models including the ENN, extreme learning machine, and long short-term memory network and the ensemble model AdaBoost.RT-ENN, which proves the effectiveness of the data decomposition and adaptive boosting algorithm.

The effects of air pollution, amongst the greatest problems facing our planet, are felt throughout all spheres of society, including the transportation sector. Its impacts can range from increased risk of illness to rising temperatures. One of the essential situations for improving inner-city general health and assisting in the creation of a sustainable environment is the ability to forecast air pollution concentrations with accuracy and effectiveness. The backpropagation model is employed in this research to predict the future concentration levels of PM2.5. Data on air quality are gathered and used in the experiment. Empirical Wavelet Transform (EWT) is used to break down the air quality data, which is then utilized to train and evaluate the model. 30% of the data collected is utilized during testing, and 70% was used to train the BP model. The evaluation criteria are applied after testing to determine the model's correctness. The Mean Absolute Error (MAE), Mean Absolute Percent Error (MAPE), and Root Mean Square Error (RMSE) are the evaluation criteria used and their values were 0.0896, 0.8112, and 0.1162, respectively.

As the main vibration source of metro depot, the throat area has seriously affected the work and life of the occupants in over-track building. However, there are few researches on vibration mitigation of turnout. Aiming at the No. 7 single turnout with 50 kg/m rails in ballast bed, the design and effect predictions of the combined vibration mitigation with under sleeper pads (USP) and under ballast mats (UBM) is carried out through the stiffness homogenization, dynamics calculation and three-degree-of-freedom vibration calculation. The study results show that: the USP not only helps to improve the turnout stiffness irregularities longitudinally, and can significantly improve the vibration mitigation effect. When using the UBM alone, the vibration mitigation effect is optimal up to 9 dB. However, after laying the USP, the vibration mitigation effect can reach about 15 dB, which increased by 67%, and meet the vibration mitigation need in the metro depot.

In order to improve the accuracy of ship trajectory prediction in waters with complex traffic conditions such as inland rivers and ports, and solve the limitation of a single LSTM in extracting time series feature information, a ship trajectory prediction model based on generative adversarial network and attention mechanism (AGAN) is proposed. The ship's trajectory is predicted collaboratively, the ability of the model to extract key information in the trajectory is improved through the attention mechanism, the relative motion information between multiple ships is extracted through the pooling layer, the individual information and the global information are fused, and finally the generative adversarial network (GAN) is used. Features that are continuously optimized in adversarial improve the accuracy of the model. The final experimental results show that the ship trajectory prediction model based on generative adversarial network has higher accuracy.

Wind farm is the carrier of wind power generation and new energy application, and the operational reliability of its power generation equipment is very important to optimize the international energy pattern and promote the achievement of China's double carbon target. This paper introduces the development status of onshore wind power, studies the power generation equipment in onshore wind farms, analyzes the indicators such as failure duration and failure rate, explores the typical failures in its operation, analyzes the key factors affecting its reliability, including gearbox, pitch system, converter, communication system and yaw system. And it puts forward the future research direction.