最新刊期
卷 23 ， 期 3 ， 2024
- 摘要：Real time news comments have the characteristics of short text， rich information， and complex structure， making it difficult for sentiment analysis to accurately capture their true emotional tendencies. To enhance semantic feature information， reduce model overfitting problems， and improve the accuracy of news comment sentiment analysis， a news comment sentiment analysis algorithm is proposed that integrates BERT model， Transformer Encoder， and multi-scale CNN model. Firstly， in response to the short length of news comments and the high content of expressing emotional views， a BERT model is used to pre train news comment texts and obtain feature vectors with contextual information； Secondly， to solve the problem of model overfitting， a layer of Transformer encoder is added downstream of the BERT model； Finally， a four channel dual layer CNN model is used to improve the performance of analyzing news comment sentiment by combining convolutional kernels of different sizes. The experimental results show that the accuracy of this method on two news comment datasets reaches 93.0% and 96.4%， respectively； The comparative experiments with different models further demonstrate the effectiveness of the proposed method.关键词：sentiment analysis;BERT model;Transformer Encoder;multi-scale CNN;news commentary
- 7
- |
- 1
- |
- 0
发布时间：2024-04-03 - 摘要：Distributed denial of service （DDoS） attacks can attack， intrude， and destroy Internet of Things devices.During the COVID-19 period， the use of a large number of IoT terminal devices for epidemic prevention and control has accelerated the frequency of information exchange， and the overly simplistic network security defense method has also made network security issues a hot topic. Deep learning （DL） has been widely used in network security to detect and respond to various network environments with low security levels. For intelligent terminals with simple structures， traditional DL models require high computing and memory resources， and often require additional operating costs when dealing with large traffic attacks. The research proposes a model based on self attention mechanism and lightweight convolutional neural networks （Self-attention-LCNN）， which extracts features from data packets within a given time period on a stream basis to detect and prevent DDoS attacks against intelligent terminals in complex network environments. The accuracy of the Self-attention-LCNN model on the CICDDoS 2019 dataset is 99.21%. Deploying the model on a raspberry pie yielded an average detection rate of 93%， indicating that the Self-attention-LCNN model has a good recognition effect in attack detection on resource-constrained intelligent terminals.关键词：DDoS;attack detection;convolutional neural network;lightweight;self attention mechanism;intelligent terminal
- 7
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Prediction of material calculation run time plays a crucial role in improving job scheduling efficiency and new material research and development. Traditional cluster job run time prediction models have poor accuracy and low availability in the field. To this end， a job prediction model based on gradient boosting decision tree is proposed， which combines domain knowledge and relevant literature to clean VASP job log data， evaluates the importance of selected features， and then conducts experiments under different data sizes and sample distribution conditions. The model is compared with models using traditional machine learning methods. The experiment shows that the average absolute percentage error of the proposed method is lower than that of traditional machine learning methods under different conditions， and the prediction error of comprehensive job running time is 4.28%， which is better than the RunningNet method's 10.3%. This proves that the proposed model has higher accuracy in predicting material calculation running time， and has a better effect on improving job scheduling efficiency and accelerating new material research and development.关键词：material calculation;job run time forecasting;decision tree;VASP job;job scheduling
- 8
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：In the information era with a large amount of information， how to efficiently and accurately retrieve the information we required is a huge challenge. Topic Crawlers are an effective way to get information about a particular domain. General topic similarity computation is based on the word granularity level， while ignoring the expression of the whole semantic feature， which will lead to the impact of both precision and recall of the crawler system. In order to solve this problem， a topic crawler method based on BTM and TextCNN is proposed， and the content topic discrimination module is considered as a text classification problem. The text semantic information is enhanced by fusing the text topic vector from BTM and Word2vec word vectors. This method uses convolutional neural network to improve the accuracy of discriminant module， which can improve the problem of inadequate representation of text features of convolutional neural network. The experimental results show that the average classification precision of the test sets in the open source news text classification dataset （THUCNews） and the real paper dataset is respectively 93.7% and 91.3% on the fused BTM and TextCNN models， which respectively increases 0.6 and 1.3 percentage points compared with the TextCNN benchmark model.关键词：topic crawler;topic similarity;TextCNN;BTM;Word2vec
- 5
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Fake reviews in e-commerce platforms mislead consumers' purchase decisions， and damage the rights and interests of consumers and legitimate businesses. The existing methods are difficult to find the implicit association between fake reviews. In order to improve the classification accuracy of fake reviews detection， a fake review detection method based on the TrustRank algorithm and graph neural network was proposed. Firstly， the features associated with the fake reviews were constructed， and the importance scores of these features were calculated. Secondly， to improve the random sampling algorithm of GraphSAGE， the suspicious values of fake reviews were calculated by the improved TrustRank method， which combined the adaptive neighborhood sampling strategy was used to select nodes in the graph with bias and adaptive and generate the neighborhood of target nodes. Finally， Yelp data set was used to verify the proposed model. The accuracy， recall and F1 of TR-GraphSAGE model were approximately 5.86%， 15.01%， and 10.12% better on average， respectively， than LSTM， TextCNN， GCN， and GraphSAGE. The TR-GraphSAGE model can eliminate the noise that affects the prediction， ensure the quality and quantity of the neighborhood， and thus improve the quality of the associated feature representation， which provides a new method for fake review detection.关键词：fake review detection;TrustRank;graph neural network;feature engineering;GraphSAGE
- 4
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：In the practice of CI/CD， automation has become a norm in software development. GitHub introduces GitHub Actions to provide software maintainers with an automated， continuous integration workflow solution. Despite providing developers with many conveniences and the GitHub community offering many third-party GitHub Actions services， only a few projects are still in use. In order to meet the needs of developers for workflow automation and reduce non development task workload， a GitHub Actions workflow project recommendation algorithm based on implicit Dirichlet distribution （LDA） topic model and Jensen-Shannon distance is proposed. By theme modeling the README file of the GitHub Actions repository， the event description text and user input of GitHub are used as model inputs to recommend GitHub Actions services for the code repository under development. The experimental results comparing the recommendation model with the standard cosine similarity based method show that this method can effectively improve the recommendation accuracy of open-source software repositories.关键词：GitHub Actions;LDA;workflow;README;repository recommendation
- 4
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：To improve the accuracy of non-invasive blood pressure prediction models and reduce the impact of individual body differences on accuracy， a non-invasive blood pressure detection model WOA LightGBM based on Whale Optimization Algorithm （WOA） and Lightweight Gradient Booster （LightGBM） is proposed. The model first extracts the preprocessed features of the photocapacitive product pulse wave and electrocardiogram signal， and combines them with human body features to form an input feature matrix； Then， the input feature matrix is dimensionally reduced using kernel principal component analysis to reduce redundancy； Finally， WOA is used to optimize the parameters of the LightGBM model. The experimental results show that the average absolute error of the WOA LightGBM model in predicting systolic and diastolic blood pressure meets the standard set by the American Association for the Advancement of Medical Devices （±5mmHg）， which has certain advantages compared to traditional models and high consistency with traditional mercury meters for measuring blood pressure.关键词：non-invasive blood pressure detection;human characteristic;KPCA;WOA;LightGBM
- 15
- |
- 2
- |
- 0
发布时间：2024-04-03 - 摘要：Mobile energy storage systems （MESS）， as emergency power sources， can play a significant role in restoring power to important loads during the post disaster recovery process. Considering that the nature of the occurrence of major power outages is due to the propagation of grid risks， a strategy for siting MESS configuration points in the framework of coupling distribution and transportation networks is proposed. Firstly， this paper describes the process of grid risk propagation with the help of SIR infectious disease model， and quantifies the risk propagation ability and risk control ability of nodes； secondly， it proposes a multi-objective optimal siting model of MESS taking into account the risk control index of nodes， and solves the problem by using immune-optimization algorithm； finally， it constructs an urban distribution grid and its geographically corresponding urban transportation system， and carries out tests and comparative analysis to validate the proposed mobile emergency storage pre-disaster placement strategy. Finally， we constructed an urban distribution grid and its geographic counterpart urban transportation system， conducted tests and comparative analysis， and verified the effectiveness of the proposed pre-disaster location method for mobile emergency energy storage， and the results show that the effect of considering the impact of grid risk propagation is also not negligible in improving the emergency response capability.关键词：distribution grids;transportation networks;mobile energy storage systems;SIR modeling;risk propagation;site location
- 2
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：In order to improve the measurement accuracy of the surface profile of a 600 mm large diameter vertical plane mirror rotated 90°along the optical axis， the finite element simulation analysis of the gravity deformation of the plane mirror with five different materials under the porous support structure was carried out， and the influence of the standing time after rotation on the surface profile accuracy was explored through the experimental measurement.The results show that the axial deformation of the plane mirror surface caused by gravity under the porous support structure is less than 14 nm， which meets the accuracy requirements.The planar mirror supported by the porous structure rotated 90°along the optical axis and reached a stable state after standing for 1 h. The PV and RMS values of the planar mirror supported by the porous structure were only 0.337 nm and 0.081 nm different from the planar mirror results after standing for 2 h. The influence of the standing time on the planar mirror results was explored.Then， it is found that the PV value of the surface profile of the plane mirror can be greatly reduced under the condition of removing at most 6% of the outer diameter， and the PV value can be reduced by more than 50 nm. A method is proposed to improve the profile accuracy of the large diameter plane mirror by setting the buffer of the supporting structure， which can effectively improve the profile accuracy of the large diameter plane mirror.关键词：large aperture flat mirror;finite element simulation;interferometric measurement;rotation measurement;experimental verification
- 4
- |
- 0
- |
- 0
发布时间：2024-04-03 -
Sine Cosine Dynamic Interference Harris Hawk Algorithm for PCNN Parameter Optimized Image Fusion AI导读
摘要：The Harris Hawk optimization algorithm suffers from the defects that the global exploitation population distribution is not extensive in the early stage and the local exploitation is easy to fall into the lack of convergence accuracy in the later stage， a Harris Hawk optimization algorithm with positive-cosine dynamic interference is proposed. Firstly， in the preliminary global development stage， two different evolutionary strategies are used to disturb the Hawk population distribution by using cosine function and sine function respectively， so as to expand the range of the population distribution， strengthen the breadth of the initial global exploration stage of the Hawk population， and provide better conditions for the local development in the later stage. Then， in the local exploitation stage， the prey escape energy formula is curvilinearly adjusted to make the prey energy loss match more closely with the real energy loss in nature， and thus enhance the capture ability in the exploitation stage. Finally， the improved Harris Hawk optimization algorithm with sine cosine dynamic interference is optimized for the three parameters of link input， time decay coefficient， and link strength of pulse-coupled neural network （PCNN） and applied to image fusion of visible and ToF confidence maps. The improved algorithm is validated by simulation experiments using six comparison algorithms and 24 test functions. The experimental data finally show that the Harris Hawk optimization algorithm based on sine cosine dynamic interference proposed in this paper can achieve better search capability and better convergence accuracy. Through the fusion comparison experiments with other fusion algorithms， it is verified that the improved fusion algorithm has significantly improved the fusion effect than the original algorithm.关键词：Harris Hawk optimization algorithm;dynamic interference;escaping energy;pulse coupled neural network;image fusion- 12
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：To solve the service composition problem of complex manufacturing tasks in cloud manufacturing environment and promote the healthy development of cloud manufacturing mode， a multi-objective service composition optimization model is constructed that comprehensively considers the interests of service demanders， cloud manufacturing platforms， and service providers. Based on the reverse learning mechanism and multi-objective population adaptive evolution mechanism， the NSGA - III algorithm is improved and applied to solve the multi-objective service composition optimization model. By comparing the mean and variance of fitness in various directions of the NSGA - III algorithm with the improved NSGA - III algorithm， the effectiveness of the latter in solving multi-objective service composition optimization problems was verified.关键词：cloud manufacturing;multi objective optimization;reverse learning;adaptive evolution;NSGA-Ⅲ
- 11
- |
- 1
- |
- 0
发布时间：2024-04-03 - 摘要：In distributed storage systems， there is often an imbalance in data access. Most access is concentrated on a small portion of the data， which can cause the system's load to be unbalanced， leading to some high load nodes becoming the performance bottleneck of the entire system. In response to the system load imbalance caused by this issue， this article proposes a dynamic local repair code （ALRC） for load balancing. This encoding scheme establishes the priority of hot and cold data through historical heat， and dynamically adjusts the layout of hot and cold data through （r， t） - availability properties， allowing the data to have additional access paths， thereby improving the parallel access performance of hot data， system load balancing， and taking into account certain storage efficiency. According to the simulation experiment results， it is shown that when the system load is relatively unbalanced， ALRC only needs a small amount of additional storage overhead compared to before encoding， which can improve the load balance by more than 53% compared to the original scheme， thereby helping to ensure uniform load distribution between nodes and improving the performance and reliability of the entire system.关键词：distributed storage;hot and cold data;load balancing;local repair code;parallel access
- 9
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Subgraph isomorphism problem is a classic and widely applicable NP complete problem. A hybrid branch learning strategy （SIBL） is proposed to reduce the solution time by combining solveless records and vertex degree constraint rules， in order to address the issue of high computational cost and dependence on vertex degree in precise algorithms. No solution record refers to the branch path without a target solution before each restart of the algorithm. In order to remove invalid searches， candidate vertices in the target graph with a vertex degree smaller than the current pattern graph vertex are first removed， and then vertices that appear in the no solution record are removed. Finally， descending sorting is performed based on the vertex score， with priority given to selecting vertices with higher scores. The new strategy provides two score calculation methods that utilize the upper bound descent to calculate a single vertex and vertex matching pairs， and alternately use the two scores to select branch vertices to quickly find the target solution， avoiding the local optimal problem of greedy selection. Through testing 14 220 examples from fields such as biology and imaging， it was found that SIBL is superior to the current leading Glasgow， McSplit+RL_ SI solved 10.08% and 19.88% of moderately difficult cases respectively， verifying that branch learning can effectively improve the solving efficiency of subgraph isomorphism algorithms.关键词：NP-complete problem;subgraph isomorphism problem;branch-and-bound;constraint rule;branch strategy
- 4
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Bearing fault diagnosis is a key issue in the health management of rotating instruments. However， in engineering practice， the observation data of bearings may be affected by some interference factors， including sensor quality and environmental noise. In traditional belief rule bases （BRB）， the model inference assumes that the input data is completely reliable， but unreliable observation data can reduce the accuracy of BRB. The Belief Rule Base Model with Attribute Reliability （BRB-r） provides a modeling framework and analysis method， and is an expert system that can aggregate unreliable quantitative data and expert knowledge. To improve the accuracy of bearing fault diagnosis， a new bearing fault diagnosis model based on BRB-r is proposed. Firstly， calculate the reliability of attributes based on statistical methods； Then， use evidence reasoning as the inference engine of the model； Finally， the projection covariance matrix adaptive evolution strategy （P-CMA-ES） is used to optimize the model parameters. The verification experiment results show that BRB-r can to some extent eliminate the influence of uncertain information in observation data and effectively process unreliable data， with good diagnostic performance.关键词：fault diagnosis;belief rule base;attribute reliability;evidential reasoning
- 0
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：After more than ten years， the traditional big data platform software has become increasingly mature.In recent years， the cloud-native architecture featuring containerization has become the preferred solution for infrastructure construction. Especially under the trend of integrating high-performance computing technology into the cloud-native environment， the design of a new generation of big data platforms is facing different challenges. These challenges involve job scheduling in a cloud-native environment， containerized adaptation of high-performance networks， and storage management under a storage-compute separation architecture. In response to these problems， this paper proposes a set of key technologies for a high-performance cloud-native big data platform， including the multi-mode workload containerized scheduling technology， the containerized RDMA data exchange technology， and the cloud-native storage-compute separation architecture. And on this basis， developed the OMBD big data platform. OMBD can effectively adapt to the characteristics of high-performance cloud native environment， and realize effective scheduling and efficient execution of multi-mode big data jobs in containerized clusters with high-performance network cards with systematic technical solutions. Experimental data and real-world application results prove that OMBD is a practical and efficient production-grade big data platform.关键词：big data;cloud-native;high-performance computing;containerization;platform software
- 32
- |
- 0
- |
- 0
发布时间：2024-04-03 -
Design of Autonomous and Controllable Cross-platform Function Adaptation Verification Test System AI导读
摘要：In recent years， the country has promoted the development of domestic information technology industry， and the demand for independent and controllable information industry has become increasingly prominent. In order to complete the adaptation verification of autonomous and controllable operating systems， database systems and various applications in the autonomous and controllable hardware and software environment， this paper proposes the autonomous and controllable cross-platform function adaptation verification test system， which can run in a variety of domestic autonomous and controllable environments， and complete the function adaptation verification test. The whole life cycle management of the test process and test environment is carried out， and the sharing of terminal resources is realized. In the process of testing， the functional test recording script is reused， and testers can run in multiple localization environments only by recording the script once. The system covers the functions of element identification， function encapsulation， etc. The tester can complete script recording with simple operation without manually writing code， which realizes automatic "zero code" recording. The system expands automatic testing， realizes the automation of adaptation and release testing， reduces testing costs， improves testing coverage and efficiency， ensures testing accuracy， and provides some reference for the adaptation verification of domestic autonomous and controllable platforms.关键词：autonomous and controllable platform;adaptation verification;domestic software;functional testing;digital transformation- 2
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：To solve the problem of numerous types of air conditioning system equipment and inconsistent protocols， an intelligent air conditioning monitoring system based on distributed monitoring and centralized management mode and B/S architecture was designed to meet the monitoring needs of the air conditioning system. Firstly， in the device monitoring layer， multithreading technology is used to collect and process real-time data from the device. Multi protocol conversion technology is used for unified protocol conversion and encapsulation， and multiple protocol data is uploaded and stored on the database server. Then， a fuzzy adaptive PID controller is used to poll and calculate the monitoring data， and the frequency converter is adjusted to control the temperature of the air conditioning system in real time. Finally， accessing the database server through the web service port enables system integration and data exchange， establishing an integrated platform for centralized control， management， and operation， achieving unified monitoring and scheduling of system equipment， unified deployment and high availability of the system， completing the interface visualization of the air conditioning monitoring system， and improving the intelligence level of the air conditioning system. Practice has shown that the system operates stably， can monitor the operation status of air conditioning in real time， and has high scalability and management efficiency.关键词：air conditioning monitoring system;B/S architecture;distributed;multi-protocol conversion;fuzzy adaptive PID
- 10
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Low noise， high contrast images can greatly improve the accuracy of doctors' diagnosis of disease. In order to solve the problem that speckle noise is introduced into the acquisition and transmission of breast ultrasound images， which leads to the deterioration of image quality and affects the early diagnosis of breast cancer， a denoising algorithm for breast ultrasound images based on improved BM3D is proposed. Firstly， a DBSCAN based superpixel segmentation method is introduced to segment the original image to obtain the corresponding superpixel label matrix； Then， using the super pixel label matrix to guide the block matching process in the BM3D algorithm can reduce the search time of the blocks to be matched， and on the other hand， the super pixel label also provides auxiliary information for similar block metrics， improving the accuracy of block matching； Finally， the hard threshold filtering in the BM3D algorithm is improved， and adaptive noise parameter estimation further improves the denoising effect. Experimental results show that the equivalent number of views of the improved BM3D algorithm is 1.75% higher than that of the traditional BM3D algorithm， and the edge retention index is 2.56% higher， while the processing time of the algorithm is reduced by 51.26%. The improved BM3D algorithm is a practical method that takes into account both noise removal effects and runtime.关键词：image denoising;BM3D;DBSCAN;adaptive filtering;breast ultrasound image
- 7
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：The automatic tumor segmentation of PET-CT images through intelligent learning methods is an important research field to assist doctors in formulating diagnosis and treatment plans. PET-CT images have the advantages of both PET and CT modalities. Traditional methods mostly simply register and fuse the images of the two modalities to extract features， ignoring the irregular tumor boundary contour of neuroblastoma. To this end， a two-stage automatic segmentation framework structure model is proposed. Firstly， use 3D convolutional neural networks to locate the tumor location； Then generate multimodal point cloud data near the segmented tumor area and extract the shape contour features of the tumor； Finally， the features extracted by the two networks are fused to predict the final segmentation result. The proposed model was compared with other multimodal methods on both proprietary and public datasets， and the experimental results verified the superiority and effectiveness of the proposed model， which can provide reference and inspiration for researchers studying the segmentation of neuroblastoma.关键词：deep learning;point clouds;multimodal;medical image segmentation;PET-CT
- 9
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：When using the classic HS optical flow method to recognize and detect moving targets in video images， there are problems such as high environmental noise and low detection efficiency. To this end， improvements were made to the optical flow method. Firstly， design a new judgment method to reduce the number of iterations for solving optical flow and the execution time of the algorithm； Then， combining edge detection and other algorithms， design algorithms that meet accuracy constraints to reduce the impact of environmental noise； Finally， GPU was used to optimize the optical flow method and edge detection algorithm in parallel， thereby further improving the efficiency of the algorithm. The experimental results show that the proposed algorithm has a detection accuracy of 88.1% on the CDNet2014 dataset， and the detected moving targets have high clarity. In addition， the maximum acceleration ratio of this algorithm on GPU is 89 times， which greatly improves its performance compared to traditional algorithms.关键词：optical flow;moving object detection;edge detection;CUDA;parallel optimization
- 15
- |
- 2
- |
- 0
发布时间：2024-04-03 - 摘要：Fine-grained image classification refers to the finer-grained subcategory division based on the divided basic categories. Due to the data features of small inter class differences and large intra class differences in fine-grained image classification， it has become a very challenging research task. Through the analysis and research of existing fine-grained image classification algorithms and models， a weakly supervised fine-grained classification method based on convolution-enhanced multi-scale feature semantics is proposed. This method correlates high-level and low-level features through convolution， uses high-level feature semantics， highlights underlying meaningful features， suppresses low-level features with invalid semantics， and obtains multi-scale features with more expressive capabilities. Based on the ResNeXt-101 network as the backbone network and the feature extraction network， the method is verified experimentally on three commonly used fine-grained image datasets， and the classification accuracy rates are 88.3%， 93.7% and 94.3% respectively. Experimental results show that this method achieves better classification results than other mainstream fine-grained classification algorithms such as the semantics method （SEF） which enhances the sub-features of global features， and multi-layer feature fusion method （MFF） which uses parallel convolution blocks.关键词：fine-grained image classification;deep learning;weak supervision;ResNeXt network
- 6
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：In order to give full play to the application of remote sensing software such as ENVI and GIS in the fine extraction and analysis of hyperspectral minerals， taking the old factory mining area and its periphery in Gejiu， Yunnan as the research area， the domestic GF-5 remote sensing data are used to extract the altered mineral information in the study area based on ENVI platform by spectral angle matching method and spectral feature fitting method. Based on statistical analysis software and GIS platform， the extracted altered mineral information in the study area is comprehensively analyzed by mathematical statistics， correlation and geophysical and geochemical remote sensing. The study shows that the alteration anomaly information extracted from the domestic GF-5 data is in good agreement with the geological and geophysical and geochemical information in the study area. The research shows that the domestic GF-5 data can be used to extract and analyze the altered minerals in the study area based on ENVI and GIS software platforms， and provide reference for the traditional geological prospecting work in the area.关键词：domestic GF-5 remote sensing data;ENVI;spectral angle matching method;spectral feature fitting method;alteration mineral information;GIS
- 4
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Pathological image cell detection is a fundamental part of medical diagnosis， and accurate detection of targeted cells and their quantities is crucial for disease diagnosis and treatment. Traditional medicine uses manual microscopy to estimate pathological images， relying on the work experience of pathologists， which leads to subjectivity and low detection accuracy. To this end， an improved YOLOv5 noise label detection and automatic correction network is proposed to detect target cells in pathological images. By using Conf and IOU functions， the network has the ability to distinguish between truth labels and noise labels， thereby achieving automatic correction of noise labels to assist doctors in clinical diagnosis of sinusitis disease types. The results showed that the improved network achieved an average accuracy and recall rate of 88.9% and 95.6% respectively on the pathological image dataset of sinusitis， which can meet the requirements of detecting pathological images and correcting noise labels.关键词：digital pathological image;unsupervision;noise labels;deep learning;self-correction
- 2
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Multi target tracking algorithms often suffer from inaccurate target recognition and poor tracking performance in complex scenes. In environments with a large number of tracked targets and severe mutual occlusion， the phenomenon of target loss is more pronounced. To this end， a multi-target tracking algorithm based on Transformer cross spatial feature association is proposed， which utilizes the advantages of Transformer structure to extract global features and multi head attention mechanism to improve the extraction ability of target features. In addition， to solve the problem of tracking target loss caused by mutual occlusion between tracking targets， a mutual attention mechanism is used to map the features between tracking targets and interfering targets for enhancement and suppression， in order to improve the accuracy and reliability of the tracking algorithm； At the same time， the degree of feature enhancement and suppression is determined based on the feature similarity between the tracking target and the interfering target. Experiments were conducted on the MOT16 and MOT17 datasets， and the proposed algorithm achieved multi-target tracking accuracy of 58.81% and 60.05%， respectively， with better performance compared to other mainstream algorithms.关键词：multi-object tracking;Transformer;feature association;object detection;attention mechanism
- 10
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Aiming at the problem of long distance from the flock of birds in the air， inconspicuous features and difficult identification in the application of airport bird repelling， a full-dimensional dynamic convolution recognition algorithm for the characteristics of birds in the air is proposed， which uses the dynamic K-value detection K-Means++ algorithm to cluster the target samples in the data set， obtain anchor frames that are more in line with different target scales， and improve the accuracy of multi-target localization and image segmentation. The full-dimensional dynamic convolution module is introduced into the backbone network of general YOLOv5s object detection and recognition， and the dynamic convolutional layer automatically adjusts the size and shape of the convolution Kernel when extracting features to adapt it to the characteristics of different birds， and makes the data more representative by dynamically convolutioning the extracted bird features. Aiming at the multiple feature maps generated by the input image after multiple convolution and pooling operations， the coherent integration is used to separate different feature maps and screen and cut off the feature channels with less obvious feature differences， thereby reducing the amount of information that needs to be calculated， thereby optimizing the detection accuracy and computational complexity of the YOLOv5s algorithm， reducing the amount of information that needs to be calculated， realizing the lightweight identification network， solving the problem that it is difficult to extract small target features such as aerial bird flocks， and improving the accuracy of bird identification.关键词：aerial bird flock;YOLOv5s target detection;full-dimensional dynamic convolution;coherent integral
- 6
- |
- 0
- |
- 0
发布时间：2024-04-03 -
Teaching Reform and Practice of the Embedded System Course under the Background of New Engineering AI导读
摘要：According to the practical needs of the cultivation of compound and engineering talents of artificial intelligence specialty under the background of new engineering， and based on the professional curriculum system and professional talent training objectives， this paper carries out the teaching reform research of embedded system course of artificial intelligence major， Some measures and methods to improve the practical teaching effect of embedded system are proposed， including the construction of creative practice curriculum system of “New engineering”， the construction of creative practical teaching platform of industry and education integration， the multi-level curriculum experimental content， the use of case teaching methods to improve teaching and learning effects， and the construction of ideological. The curriculum reform takes the cultivation of engineering ability as the new requirement， student interest guidance as the new approach， industry development needs as the new concept， and explore the practical ways of “new engineering” professional talents' engineering innovation ability training mode， so as to realize to the professional talent training objectives of embedded system practical courses in artificial intelligence major.关键词：embedded system;project-based learning;teaching reform;case teaching;online and offline mixed teaching- 6
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：In the process of collaborative knowledge construction， learners often only care about the expression and improvement of their personal opinions， and lack the sense of community to communicate and negotiate with their peers， which weakens the group cohesion to a certain extent and hinders effective interaction with their peers. In response to this problem， this research introduces the group dynamics theory that has a significant impact on group cohesion into the teaching activities of collaborative knowledge construction， combines the circle of influence factors of group cohesion with the knowledge improvement circle of collaborative construction， and follows the design research based methods and ideas， designed and implemented collaborative knowledge construction learning activities that promote group cohesion for comprehensive practical courses in junior high schools. Through data analysis， it is known that this method can effectively promote learners' interaction enthusiasm and the level of collaborative knowledge construction.关键词：group dynamics;collaborative knowledge construction;group cohesion;activity design
- 7
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：In response to the characteristics and difficulties of advanced language programming courses， combined with the educational philosophy of OBE （Output based education）， the goal is to improve students' practical application ability and comprehensive quality. Reform the teaching of courses from multiple aspects such as ideological and political education， teaching models， teaching design， and comprehensive evaluation methods. While improving teaching quality， we also focus on cultivating students' ability to solve practical problems， helping them establish a correct outlook on life and values， and providing more motivation and support for subsequent learning.关键词：OBE;advanced language programming;curriculum ideology and politics;teaching design
- 8
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Mobile teaching tools are important intelligent tools for modern teaching， and analyzing their characteristics and development trends is beneficial for their rational application. A total of 358 articles related to the theme of mobile teaching tools in core journals indexed by China National Knowledge Infrastructure （CNKI） from 2007 to 2021 were selected as the research objects. Quantitative methods were used to analyze their quantity distribution， core authors， research institutions and source journals. At the same time， time zone distribution， high-frequency words， clustering maps， and emergence analysis were performed on the keywords of the articles， Strive to explain the following questions： ①What kind of historical development and evolution has mobile teaching tools gone through？ ②What are the hotspots and characteristics of mobile teaching tools in current practice？ ③What are the trends in the research field of mobile teaching tools？ The analysis results show that the research on mobile teaching tools in China can be roughly divided into three stages： the budding period of mobile teaching tools emerging in response to the trend， the rapid development period presenting diverse characteristics， and the transformation period of exploring the construction of smart integrated new platforms. The current research on mobile teaching in China mainly focuses on four aspects： educational theory and reform momentum， technology application and platform development， mode exploration and teaching design， teaching services and learning support. Looking ahead to the future， integrating deep teaching concepts and deep learning theories is an effective path to deepen the application of mobile teaching tools and return to the essence of education.关键词：mobile teaching tools;knowledge graph;bibliometric analysis;deep learning
- 9
- |
- 0
- |
- 0
发布时间：2024-04-03 - 摘要：Steel is an indispensable raw material in the industrial field， and surface defects seriously affect the quality of steel. Traditional steel surface defect detection methods have low accuracy， slow speed， and high labor intensity， which cannot meet actual production needs. In recent years， deep learning technology has developed rapidly， which can fully explore the underlying feature information of target images， bringing new solutions to steel defect detection. Summarize the relevant literature on steel surface defect detection methods in recent years， briefly describe the principles and applicability of traditional detection methods， analyze the structure and characteristics of deep learning detection models， and summarize some technical difficulties in the current field， and look forward to future development trends.关键词：steels;surface defects;target detection;deep learning
- 29
- |
- 3
- |
- 0
发布时间：2024-04-03 -
Subject Knowledge Graph： Connotation， Technique Architecture， Application and Development Trend AI导读
摘要：As an artificial intelligence technology， knowledge graph is one of the key supporting technologies for the development from perceptual intelligence to cognitive intelligence. Its application in the field of education will undoubtedly promote the innovative development of intelligent education. Firstly， from the perspective of the development process of knowledge graph technology and discipline， the technical and knowledge essence of disciplinary knowledge graph is examined， and the value positioning of disciplinary knowledge graph is clarified. Secondly， based on the analysis of the evolution and development of disciplinary knowledge graph construction technology and models， the basic functional architecture and technical system of disciplinary knowledge graph are constructed， and its typical application scenarios in intelligent education are analyzed. Finally， from the perspectives of application scenarios and technological ontology development， this paper provides a prospect for the development trend of disciplinary knowledge graphs， in order to provide reference for the application of disciplinary knowledge graphs in intelligent education.关键词：subject knowledge graph;technique framework;development trend- 2
- |
- 0
- |
- 0
发布时间：2024-04-03
- Technical support is provided by Beijing Founder electronics co., LTD 鄂ICP备12016542号-70
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0