最新刊期
卷 22 ， 期 4 ， 2023
- 摘要：The ability to extract entities and relationships from unstructured text is crucial for automatically maintaining large-scale knowledge graphs. However， existing joint extraction methods cannot effectively handle the problems of entity pair overlap （EPO） and single entity overlap （SPO） in triplet extraction. To this end， a joint extraction model based on entity pair classification is proposed， which directly decodes fact triplets using the boundary span between the classified head and tail entities. Firstly， the model uses a pre trained language model to encode text and obtain a token embedded representation of the text； Then， enumerate and integrate token pairs， and use softmax to predict the labels of token pairs； Finally， based on relationship and label prediction， the head and tail entities decode the corresponding fact triplets for the token span. Experiments have shown that this method can fully capture the interdependence of various elements in a triplet and predict entity pairs and their relationships at the same time. It has shown good performance in solving problems such as relationship overlap and triplet element dependency. The comprehensive performance index （F1 score） of the model on the NYT dataset reaches 92.1%， verifying its effectiveness in entity relationship joint extraction tasks.关键词：joint extraction;entity relationship extraction;triples;token pairs
- 123
- |
- 34
- |
- 0
发布时间：2023-05-05 - 摘要：In the field of time series classification， Shapelets-based methods are interpretable as they can target the key parts for classification. However， high time consumption and poor performance compared with deep learning methods impedes its application. While deep learning methods could achieve better performance， they are lack of interpretability. To improve the performance of Shapelets-based methods and enhance the interpretability of deep learning models，we proposed a new method called Kernel-Shapelets. Based on the weight corresponding to kernels of CNN and the output of Global Max Pooling Layer， it could generate shapelets and locates the key basis of CNN for classification. Through experiments on the UCR time-series dataset， the average classification accuracy of the Kernel-Shapelets is 82%， improving 15.4% compared to the optimal benchmark Shapelets-based model. The results demonstrate that Kernel-Shapelets can extract more effective and discriminative Shapelets by using the learning ability of the CNN network， while improving the interpretability of CNN.关键词：time series data mining;time series classification;deep learning;CNN;Shapelets
- 76
- |
- 21
- |
- 0
发布时间：2023-05-05 - 摘要：Aiming at the problem that the current automatic modulation recognition （AMR） algorithm has poor classification performance and is difficult to distinguish high-order modulation signals， propose a collaboration model CLCM that combines convolutional neural network （CNN） and long short-term memory （LSTM） to complete the classification task by voting combined strategy. The experimental results show that the recognition accuracy of CLCM exceeds that of most current advanced models （such as XGBoost， VGG， etc.） by more than 20% when the signal to noise ratio is over 12 dB. It can effectively identify 24 widely used modulation signals （the average recognition accuracy is higher than 95%）， and the recognition accuracy of high-order quadrature amplitude modulation （QAM） signals （such as 16QAM、64QAM、256QAM） is over 90%.关键词：modulation recognition;CNN;LSTM;deep learning;wireless communication
- 29
- |
- 5
- |
- 0
发布时间：2023-05-05 - 摘要：In recent years， because of the growing demand for location based services in indoor environment， indoor localization has attracted extensive attention from industry and academia. In order to reduce the labor cost of offline fingerprint collection and improve the accuracy of online location， an indoor fingerprint interpolation and location method based on deep autoencoder was proposed. The received signal strength （RSS） signal of the Wi-Fi access point is processed to establish an offline fingerprint database； the deep neural networks （DNN） model is used to estimate the signal intensity of other locations and generate corresponding location fingerprints， which improves the density of fingerprint database and reduces the dependence on fingerprint database. On this basis， a stacked autoencoder （SAE） is used to high level extracted features from RSS signals to improve localization performance. Experimental results on public datasets and self-collected datasets show that the proposed method has high localization accuracy in both region class and precise class； by adding newly generated data and combining with the proposed positioning method， the positioning performance is improved and the robustness of the system is enhanced.关键词：indoor localization;fingerprinting interpolation;autoencoder;deep neural network
- 30
- |
- 8
- |
- 0
发布时间：2023-05-05 - 摘要：Due to the large amount of Internet data， in order to efficiently manage the massive Chinese text data on the Internet， a Chinese text classification method based on Albert and TextCNN （referred to as ATT） is proposed. This method introduces Albert model to solve the problem of polysemy. TF-IDF algorithm is used to extract the five words with the highest weight in the current text data to build the whole document keyword table. The keyword table and the word vector generated by Albert are vector spliced to form a polysemy word vector that integrates keyword information. Moreover， on the basis of traditional TextCNN， the window size of convolution kernel is adjusted according to the characteristics of Chinese language to extract the deep local features of text data. The experimental results show that the ATT model， compared with the traditional model without TF-IDF keyword list and without adjusting the size of convolution kernel， increases the F1 value by 1.88% and 2.26% respectively， providing a new method for Chinese text classification.关键词：word vector;text feature extraction;multi-label;text classification
- 22
- |
- 10
- |
- 0
发布时间：2023-05-05 - 摘要：Tax regulations are an important type of tax data in tax laws， serving as the legal basis for strengthening tax collection management and regulating tax collection and payment behavior. Firstly， in response to the complex knowledge， numerous professional terms， and obvious logical structure of tax regulations， a knowledge modeling method for tax regulations is defined. It is proposed to represent tax regulations as a sub graph of regulation knowledge， and fuse the taxpayer source nodes in the sub graph to form a knowledge graph representation； Then， a feature extraction algorithm is designed for the text features of tax regulations， and a BERT based model for extracting knowledge elements of tax regulations is proposed. Compared to traditional sequence annotation models based on static word vectors， the P-value， R-value， and F1 are significantly improved； Finally， based on the feature extraction model， automatic construction of a knowledge graph for tax regulations is achieved， providing semantic support for smart tax applications. Establishing a knowledge graph for tax collection regulations is of great practical significance for promoting the application of "smart taxation".关键词：tax regulation;knowledge graph;knowledge modeling;knowledge subgraph;elements extraction;smart taxation
- 12
- |
- 2
- |
- 0
发布时间：2023-05-05 -
A Bearing Fault Diagnosis Method Based on Improved One-dimensional Convolutional Neural Network AI导读
摘要：Aiming at solving the problems of low diagnostic accuracy and over fitting of the classical one-dimensional convolutional neural network model， a bearing fault diagnosis method based on improved one-dimensional convolutional neural network is proposed. The normalization method is used to increase the generalization ability of the model. For the problem of model over fitting， the dropout method is adopted. Finally， the proposed method is applied to the experimental data of rolling bearing failure of Case Western Reserve University. The experimental results show that compared with the classical convolutional neural network， the proposed method can improve greatly the diagnostic accuracy （ as high as 99.79%）. Moreover， the entire diagnosis process does not require any manual feature extraction， which can reduce the loss in the feature extraction process， and the method can realize end-to-end detection and has better versatility.关键词：fault diagnosis;convolutional neural network;deep learning;Dropout method- 14
- |
- 8
- |
- 0
发布时间：2023-05-05 - 摘要：Prediction algorithm based on CEEMDAN-SE-TCN is proposed according to the volatility and nonlinearity of server cluster load data. First， the original server cluster data is decomposed into adaptive noisy set empirical mode decomposition （CEEMDAN）， which effectively reduces the complexity of the load sequence. Then， after the decomposed relevant IMF components are obtained， the correlation coefficient method is used to compare each IMF component with the original sequence to remove the components with weak correlation. Finally， the corresponding eigenvalues of each component are extracted and input into the time convolution network （SE-TCN） with attention mechanism for modeling and prediction. The actual measurement of CPU load rate sequence in Google cluster dataset shows that CEEMDAN-SE-TCN model is better than other benchmark models under the same conditions， and MAPE indicators are reduced by 7.1%， 6.5% and 2.5% respectively compared with other models， which proves the effectiveness and feasibility of this algorithm.关键词：adaptive denoising ensemble empirical mode decomposition;correlation coefficient method;attention mechanism;time convolution network;load prediction
- 12
- |
- 3
- |
- 0
发布时间：2023-05-05 - 摘要：Big data has brought new opportunities for enterprise digital operation and management. The data contains the real emotional feedback of users， which is conducive to enterprises to timely grasp their product service quality， adjust enterprise marketing strategy and optimize product performance. It is important to improve the competitiveness of enterprises in their peers and accelerate the digital transformation of enterprises. Thus，the method of enterprise intelligent operation management based on sentiment analysis on the user comments is proposed which can help the enterprise decision makers understand existing problems and formulate its solutions through data prepossessing， data exploration， data classification modeling for user comments by big data sentiment classification technology and data visualization. Taking the competitiveness of R Hotel as an example， a polynomial Naive Bayesian sentiment classification model based on Unigram and Bigram is established. And the corresponding prototype system is developed， and the feasibility of the method proposed is verified after putting it into practice which provides a new idea and method for enterprise scientific digital operation and management.关键词：data driven;sentiment analysis;enterprise operation management;user comments;machine learning
- 17
- |
- 4
- |
- 0
发布时间：2023-05-05 - 摘要：Character identification can use single modal biometric features such as gait or sound， but this method is often limited in practical applications due to factors such as occlusion and security. To this end， a feature fusion method for character identity recognition is proposed. Firstly， GaitSet and ResNet models are used to extract gait and voiceprint features， and then attention mechanism is used to fuse the above features. Finally， raw audio data with noise interference and audio data after noise reduction are used for experiments. The results show that the feature fusion method achieved recognition accuracy of 80.27% and 83.64% respectively， both higher than the single mode method， and has better robustness in the presence of interference.关键词：gait recognition;voiceprint recognition;multimodal fusion;human identification
- 12
- |
- 3
- |
- 0
发布时间：2023-05-05 - 摘要：Classifying rock samples contributes to mineral mining and geological research， to meet the goal of real-time on-site rock classification on a lightweight platform， the VGG-T3 model based on the VGG16 model is proposed， by modifying the fully connected layers and convolutional layers of the VGG16 model， three new models are obtained， some of the VGG16 pre-trained parameters are used while transfer learning and freeze training， and subsequently comparing them to arrive at the optimal model named VGG-T3， finally， the VGG-T3 is unfrozen for normal training to obtain the final model parameters. the classification result of the rock sample by the VGG-T3 model shows that the VGG-T3 model reaches 100% accuracy rate in the training set and 97.81% accuracy rate in the verification set， meanwhile 97.74% accuracy in the test set， this model can serve as a provide reference for rock sample classification， which reached expectation.关键词：deep learning;rock recognition;convolutional neural network;transfer learning;freeze training
- 15
- |
- 2
- |
- 0
发布时间：2023-05-05 - 摘要：The human face is affected by internal and external factors such as environment， habits， genes， gender， race， etc.， resulting in accurate facial age estimation is still a very challenging topic. Therefore， propose a gender-assisted task-based dual-branch age estimation model CUT-ResNet50 to explore the impact of gender factors on the age estimation task， applying label distribution learning to age estimation. The low-level features extracted by the shared parameter layer are input into the gender network branch to extract gender features， and then fused with the low-level features of the shared parameter layer as the input of the age network branch. The label distribution learning is introduced， and the adjacent information of the dataset is used to generate the age distribution to reduce the impact of the uneven distribution of face samples， and retrain the dual-branch age estimation model. The experimental results show that on the public MORPH-II dataset and UTKFace dataset， the MAE reaches 2.1 and 4.1， respectively， which shows that multi-task learning and label distribution learning with gender auxiliary tasks can improve the accuracy of age estimation， which proves that effectiveness of our new design model.关键词：multi-task learning;attention mechanism;label distribution learning;feature fusion
- 28
- |
- 6
- |
- 0
发布时间：2023-05-05 - 摘要：In named entity recognition tasks， word-level embeddings are generally used to vectorize sentences， while ignoring the information contained in character-level features. Therefore， propose a model that combines character-level features and word-level features， uses BiLSTM to capture character-level information， and then combines the output character vector with the word vector obtained by the Glove model and outputs it to the BiGRU network. The attention layer mechanism redistributes the weights of the features， and finally outputs the label sequence with the globally optimal section by using the CRF layer. The experimental results show that the proposed method achieves better performance in the named entity recognition task than the BiGRU-CRF model. The F1 values on the CoNLL-2003 and OntoNotes5.0 datasets reach 91.69% and 88.97%， respectively，which proved the propose method helps improve the performance of named entity recognition tasks.关键词：named entity recognition;character embedding;bidirectional gated recurrent unit;self-attention;conditional random field
- 13
- |
- 3
- |
- 0
发布时间：2023-05-05 - 摘要：With the complexity and large scale of academic networks， most traditional similarity metric algorithms rely on network patterns and lack universality， while the computational efficiency is low on large-scale networks. Aiming at the above problems， propose a link-based similarity measure. First， using the global heterogeneous relationship distribution for statistical analysis and generate a relationship feature matrix to normalize the importance of different types of link relationships. Next， extend the TopSim algorithm with the relational feature matrix adapts the algorithm to the similarity metric of large-scale academic networks， while accelerating query process by threshold filtering. Finally， return top k nodes related most to each query. Results on real datasets indicate that when considering the data composition and its distribution characteristics， query result accuracy is improved by 7.25% on average with guaranteed query efficiency.关键词：academic networks;similar metric;relational distribution;relationship feature matrix;TopSim algorithm
- 18
- |
- 7
- |
- 0
发布时间：2023-05-05 - 摘要：When solving the problem of class imbalance， traditional classifiers have disadvantages such as uneven distribution of samples and large classification errors of minority classes. In order to improve the classification performance of the minority class and the whole， an adaptive heterogeneous integration model is proposed. The model uses nine sampling methods and nine classification methods， adaptively selects the base classifier with the highest AUC according to the data distribution， and then integrates all base classifiers to obtain the prediction result. During the experiment， two comprehensive performance metrics， AUC and F1-measure， were used to conduct experiments on 14 unbalanced datasets from KEEL and UCI databases. The experimental results show that the proposed model is more robust and superior when dealing with class imbalanced data. Compared with the conventional method， its average AUC value is improved by 5.98 %， and the average F1 value is improved by 5.88 %.关键词：integrated learning;heterogeneous integration;imbalanced data;classification
- 17
- |
- 5
- |
- 0
发布时间：2023-05-05 - 摘要：The quantitative characteristics of evoked EEG in schizophrenic patients can provide basis for disease diagnosis. Therefore， event related potentials were obtained from learning tasks of 42 schizophrenics and 29 healthy people for research. First， the time domain and frequency domain characteristics of human EEG signals are extracted. Then， the brain network is constructed based on the coherence and phase lag index to analyze the differences of brain functional connectivity. The experimental results showed that the P300 amplitude of schizophrenic patients was significantly lower than that of healthy people during the learning task. Within 0.1-0.6s after the start of the stimulation， the theta frequency band power of healthy people was stronger. The difference in the connectivity of brain functions between the temporal lobe and the occipital lobe of the left hemisphere was mainly reflected in the difference， which could provide reference for early diagnosis of schizophrenic patients.关键词：schizophrenia;time domain characteristics;time-frequency characteristics;functional connectivity
- 12
- |
- 4
- |
- 0
发布时间：2023-05-05 - 摘要：Designing a reasonable tourism route for Jingdezhen ceramic cultural scenic spot is conducive to improving the visibility and cultural inheritance of a city. However， traditional route planning algorithms often have defects of low planning accuracy and high time complexity in tourism route planning. Based on this， A Weighted Disturbance Mechanism simulated Annealing Genetic algorithm （WDMAGA） was proposed. WDMAGA introduces the weight disturbance mechanism of the path length change function in the path evaluation criterion， and integrates the temperature annealing factor in the simulated annealing algorithm， so that the evaluation value of each path is closer to the actual situation. At the same time， the crossover and mutation processes with simulated annealing factors were designed respectively according to the changing rules of cosine function and exponential function， so that the population fitness changed adaptively with the search process. Simulation test results show that： in the process of tourism path planning for Jingdezhen ceramic cultural scenic spot， WDMAGA algorithm can effectively and quickly plan a tourism path that is more consistent with the actual situation.关键词：scenic area tourism;path planning;genetic algorithm;simulated annealing algorithm
- 9
- |
- 3
- |
- 0
发布时间：2023-05-05 - 摘要：The filtering algorithm of text is widely used， with the rapid development of social platforms such as WeChat and Weibo， short text messages occupy the mainstream in network communication， and the filtering of short text messages is becoming more and more important. By comparing the advantages and disadvantages of classic pattern matching algorithms such as BF algorithm， KMP algorithm， AC algorithm， etc. ， this paper selects the DFA algorithm that is more suitable for short text filtering， introduces the basic principles of DFA algorithm， and proposes an improved algorithm based on DFA， which improves the detection rate through sensitive word preprocessing and filtering process optimization. Experimental results show that the improved algorithm improves the accuracy of Chinese dialogue dataset detection by 3% and reduces the false alarm rate by 0.87%， which has high application value.关键词：short text;message filtering;DFA;pattern matching;Trie tree
- 26
- |
- 2
- |
- 0
发布时间：2023-05-05 - 摘要：In the long-distance transmission of digital signals， the high-order PSK modulation signal has the problem of improving the signal propagation quality at the expense of power utilization. The main types of multi-ary digital modulation in the P-band are Multi-ary Amplitude Phase Shift Keying （MAPSK） and Multi-ary Quadrature Amplitude Modulation （MQAM）. At the same time， there are unknown signals in the P-band， such as single tone signals and voice signals， which need to be distinguished. Common modulation type recognition algorithms often use a single-stage receiver to extract a single feature to distinguish modulation types， the method of distinguishing features extracted by a single-stage receiver has poor reusability， and the single feature extracted has poor recognition performance. To solve this problem， a recognition algorithm of amplitude phase mixed modulation signal based on the combination of multi-level blind digital receivers is proposed. Taking the traditional decision tree as the framework， the receivers with different phase discrimination methods are used to extract the signal amplitude， phase and frequency information in different decision branches. The extracted information is matched with various types of feature templates to count the signal feature results step by step， and the signal type is judged by the feature results. The signal recognition rate of the algorithm is more than 90% when the signal-to-noise ratio is 10dB in P-band.关键词：multi-stage digital receiver;carrier synchronization;amplitude and phase mixed modulation signal recognition;template matching
- 5
- |
- 2
- |
- 0
发布时间：2023-05-05 - 摘要：With the gradual deployment of IPv6 network scale， the number of IPv6 network attacks continues to increase， and there are security vulnerabilities at the system， application， hardware and protocol levels. To this end， the security risks of IPv6 are summarized from three aspects： the same security risks of IPv6 and IPv4， the unique security risks caused by the new features of IPv6， and the security risks in the transition period of IPv4/IPv6， and the prevention schemes are summarized. First， aiming at the same network attack as IPv4， the IPv6 network attack protection technology is studied. Then， compared to IPv4， IPv6 headers add flow label fields， extended headers， have larger address space， stateful address configuration DHCP upgraded to DHCPv6， stateless address configuration ICMP upgraded to ICMPv6， add Neighbor Discovery Protocol， etc.， and the unique security risks and prevention schemes caused by the new features are discussed. Finally， aiming at the security risks of transition mechanisms such as dual-stack， tunnel and translation used in IPv6 deployment， the security protection scheme is studied to provide theoretical and technical support for discovering unknown attacks and effectively resisting network attacks.关键词：IPv6;IPv4;transition mechanisms;security threat;prevention scheme;network security
- 14
- |
- 2
- |
- 0
发布时间：2023-05-05 - 摘要：Under normalized epidemic prevention and control， telemedicine has become the first choice in medical institutions. The 5G network provides better network capabilities for telemedicine. At the same time， medical image also faces huge threats in network transmission， such as privacy leakage， malicious tampering， and illegal copying. To solve the problems， a reversible data hiding technique for medical images based on plaintext camouflage encryption and contrast enhancement is proposed. First， the original medical image is conducted by histogram stretching to enhance contrast and improve capacity and visual quality. Then， divide it into complex texture blocks and simple texture blocks according to complexity， and the different algorithms are performed in different blocks to hide the patient's private data. Finally， the plaintext camouflage encryption is applied to disguise the stego-image as a completely different medical image. In plaintext camouflage encryption， K-means classification information is further sorted to improve the suitability of the target image and the marked image， so that the visual quality of the fake image is better. The experimental results show that the algorithm has higher security， the plaintext camouflage encryption has a better protection effect on medical images. Furthermore， the proposed scheme provides superior outcomes in terms of PSNR， SSIM and RCE of reversible data hiding.关键词：reversible data hiding;plaintext camouflage encryption;contrast enhancement;texture division;privacy protection
- 8
- |
- 4
- |
- 0
发布时间：2023-05-05 -
Exploration of Network Security Construction in Colleges and Universities Based on CDN Technology AI导读
摘要：Due to the large number of students in the Open University， most of them are online learners. In order to ensure the network security of universities， web servers are protected through CDN combined with WAF. The combination of CDN filtering DDoS attacks and WAF blocking malicious Web attacks can effectively block network attacks that cannot be isolated by traditional firewalls. At the same time， in the construction of the school's security management system， a closed-loop management of network security is proposed， which improves the network security management system from aspects such as establishing a network security system， building a talent team， providing security operation and maintenance services， emergency response plans， security training， and "equal protection 2.0". Practice has shown that by deploying CDN services on the learning platform， students have a better learning experience and information transmission is more secure and reliable. This method can provide reference and inspiration for the construction of network security in universities.关键词：network security;CDN;closed-loop management;open university;educational informatization- 6
- |
- 3
- |
- 0
发布时间：2023-05-05 - 摘要：Recently， self-supervised models based on angular rotation are introduced into generative adversarial networks， which alleviate the problems of generative adversarial model stability and mode collapse to some extent. However， this self-supervised model is based on global information comparison， lacks local correlation， and lacks the ability of model feature extraction， resulting in high global similarity and low pixel accuracy of generated images. In this paper， a new type of combat model was generated from the supervision， the image window （patches） is introduced into the traditional angle rotation from the supervision model， through the image block， the local image and overall image as input， the rotation angle of discriminant criterion based on the new position， improve the global similarity model as well as to enhance the local characteristics of the correlation. We conducted tests on CIFAR-10 and STL-10 data sets， and the experimental results show that the self-supervised generation model based on patches is better than the angle-rotation self-supervised model.关键词：deep learning;self-supervised learning;generative adversarial nets;patches
- 8
- |
- 3
- |
- 0
发布时间：2023-05-05 - 摘要：Aiming at the problems of low recognition accuracy and large scale network models of existing crop disease image recognition methods， a crop leaf disease recognition method based on MSA-ResNet（Multi-scale Attention Residual Network）is proposed. Firstly， the multi-scale convolutional layer is used to extract different scale disease features to widen the network width； on the premise of not losing the network performance， depthwise separable convolutions is used to replace some standard convolution in the network structure to reduce the model scale； Secondly， the attention mechanism module is used to enhance the expression of effective disease features and weaken the influence of irrelevant features； Finally， multi-scale feature fusion is performed for different channels to extract rich disease features and improve the accuracy of the model. The experimental results show that the improved multi-scale residual network MSA-ResNet achieves better performance on the AI Challenger dataset， with a recognition accuracy rate of 89.93% and a model size of 30.5MB， which balances the recognition accuracy of the model and model size well.关键词：disease identification;residual network;multiscale convolution;depth-separable convolution;attention mechanism
- 22
- |
- 2
- |
- 0
发布时间：2023-05-05 - 摘要：Aiming at the problems of slow speed and low efficiency of traditional image stitching algorithm in UAV remote sensing image stitching process， which cannot meet the requirements of real-time and accurate stitching， an improved ORB image stitching algorithm is proposed. Firstly， the scale pyramid is constructed and the feature points are extracted by ORB algorithm， and then the feature points are described by BEBLID descriptor； The nearest neighbor ratio （NNDR） algorithm is used for rough matching， and then the optimal geometric constraints are constructed based on the feature point voting to further optimize the feature points. The random sampling consistency （RANSAC） algorithm is used to calculate the transformation matrix and obtain the high-precision transformation matrix； Finally， the improved gradual in and gradual out weighted fusion algorithm is used to realize image mosaic. The experimental results show that the registration accuracy of the proposed algorithm reaches 100% at the highest， the registration time is less than 0.91s， and the information entropy of mosaic image reaches 6.807 9. Compared with the traditional algorithm， the algorithm in this paper has higher splicing efficiency， and can obtain higher quality splicing images while reducing the image splicing time. The algorithm performance is significantly improved.关键词：image mosaic;multi scale FAST detection;BEBLID feature;optimal geometric constraint
- 10
- |
- 2
- |
- 0
发布时间：2023-05-05 - 摘要：The color image is grayed to reduce the image noise and the PCA_CNN network calculation. The Canny operator is used for edge processing， and the OTSU algorithm and contour extraction algorithm are used to complete the bank card digital area recognition， so that the recognition task has strong adaptability and anti-interference to the background， illumination and contrast of the image. Therefore， principal component analysis （PCA） is used to select the convolution kernel of convolutional neural network （CNN） model to avoid time and computational waste caused by a large number of iterations. The sigmoid activation function is improved to make it piecewise monotonically increasing， so as to improve the accuracy of recognition. The recognition rate of the improved PCA_CNN model is 98.53%， which has certain advantages in accuracy and convergence speed compared with CNN method， traditional BP neural network， SVM model， Bytes model and Siam network model. The experimental results show that the PCA_CNN model can effectively extract the bank card number from the bank card photos taken by non-professional photographic equipment such as mobile phones under natural light.关键词：PCA algorithm;CNN model;bank card number recognition;deep learning
- 15
- |
- 2
- |
- 0
发布时间：2023-05-05 - 摘要：At present， most image super-resolution networks improve the reconstruction ability by deepening the convolution neural network layers and expanding the network width， but greatly increase the model complexity. To this end， a lightweight image super-resolution algorithm is proposed. Through the two-branch feature extraction algorithm， the network model can be fused and output the feature information of different scales at one time， and the pixel attention branches are combined to add weights to each pixel respectively， which only enhances the feature expression of pixel details at the cost of fewer parameters. In addition， the up-sampling part combines subpixel convolution and neighborhood interpolation methods to extract feature depth and spatial scale information respectively， and output the final image. In addition， the subpixel convolution integral branch of the combined attention mechanism further strengthens the important information and makes the output image have better visual effect. The experimental results show that the model achieves similar reconstruction performance to the CARN model with a parameter quantity of 1 592K when the parameter quantity is only 351K， and the SSIM value in some test sets is higher than the CARN value， which confirms the effectiveness of the proposed method and can provide a new solution for lightweight image super-resolution reconstruction.关键词：image super-resolution;lightweight;pixel attention;multi-scale upsampling;image processing
- 39
- |
- 6
- |
- 0
发布时间：2023-05-05 - 摘要：The research on teaching contents and teaching methods of university AI cross-applications has a key role in promoting the development of AI and the improvement of university students' ability to solve AI-related application problems. To address the problem of teaching content and teaching method design of AI cross-application in university， introduce the demand of AI cross-application， AI technology system， related application fields， and the difference of teaching objects， and analyzes the relationship between these contents and AI cross-application talents training； at the same time， combining two specific examples of AI cross-application， it elaborates the teaching content of AI cross-application courses for university students. At the same time， the teaching contents covered by the AI cross-application course for college students are explained with two specific examples of AI cross-application； finally， the teaching method of integration of teaching， learning， application and creation is proposed with our own teaching practice experience， and the effectiveness of the teaching method is analyzed from the perspective of students' employment and further education.关键词：artificial intelligence;cross-application;pedagogical content;teaching method;talent development
- 13
- |
- 4
- |
- 0
发布时间：2023-05-05 - 摘要：Under the requirements for the new engineering construction of specialized courses， the problems in teaching computer network courses have been analyzed in depth from teaching content， the relevance of teaching mode， teaching method， examination mode and the ideological and political curriculum. The teaching system of computer network courses has been reconstructed with the guidance of problem-driven and ability cultivation. It mainly optimizes the construction of teaching content that integrates teaching and practical applications， online and offline teaching methods， the way to improve teaching through competition， and the way for ideological and political teachers to promote ideological and political construction and open assessment of the curriculum. Let students not only master the basic computer network knowledge but also cultivate their social responsibility and moral character and promote the transition of students' professional ability to vocational ability.关键词：emerging engineering;computer network;teaching system;ideological and political curriculum
- 16
- |
- 6
- |
- 0
发布时间：2023-05-05 - 摘要：Curriculum ideological and political education is a requirement of national and university education. In order to explore new teaching methods of curriculum ideological and political education， identify the entry point， and effectively integrate curriculum ideological and political education into the classroom teaching of program design principles， deeply explore how to carry out curriculum ideological and political reform from the perspectives of curriculum teaching cases， classroom teacher-student interaction， and after-school training programs，and propose and practice innovative reform programs such as case teaching， discipline competitions， and curriculum ideological and political elements， so that students can improve their understanding of ideological and political content while learning professional courses， and shape their scientific and craftsman spirit. The evaluation results show that the curriculum reform has not only achieved significant results， but also provided rich experience for the ideological and political reform of similar courses， with a view to providing reference and reference for the ideological and political teaching reform of courses.关键词：programming principle;ideological and political education;reform in education;case teaching
- 13
- |
- 7
- |
- 0
发布时间：2023-05-05 - 摘要：With the continuous promotion of the construction of national first-class undergraduate majors， it is essential to explore the cultivation of first-class undergraduate majors with bachelor's degrees in local colleges and universities. The graduation thesis is strictly guided by the graduate thesis system and the graduate thesis system； Establish a three-dimensional， whole process and four-track education model； Improve the professional curriculum system by closely following the development and cross-application of new technologies； Build practical teaching projects that go deep into the industrial chain and innovation chain， create practical teaching gold courses updated with new technologies， and promote the reform of professional practical teaching； Promote the talent training of first-class undergraduate majors in local colleges and universities by integrating the training mode of first-class undergraduate majors in colleges and universities， enterprises， governments， local governments， and industries. Students have made promising innovations， application abilities， and actual graduation thesis achievements through reform and practice.关键词：first-class undergraduate major;outcome orientation;double tutorial system;applicantion-oriented;coordinated education
- 4
- |
- 3
- |
- 0
发布时间：2023-05-05 - 摘要：To improve the teaching quality of the website principle and design courses， research-based on a task-driven mode under the concept of outcomes-based education （OBE） is carried out. First of all， the graduation requirements that support the achievement of the training objectives and the curriculum ability objectives are determined. Secondly， the task-driven teaching content-oriented by the learning outcomes is designed. In addition， a diversified assessment and an effective evaluation mechanism are established. Finally， The implementation effect is analyzed through teaching practice， the effectiveness and sustainability of the teaching mode are confirmed.关键词：OBE;website principle and design;curriculum ability objectives;task-driven mode;diversified assessment
- 8
- |
- 2
- |
- 0
发布时间：2023-05-05 - 摘要：Recently， with the popularity of live culture and the rapid development of “live + education”， live classroom has become a new form of online learning. The live classroom provides teachers and students with effective support for real-time teaching interactions such as knowledge teaching， problem exploration， and emotional communication， which makes classroom teaching more effective. However， the research found that the current live classroom mainly focuses on some basic interactions such as teaching， counseling， discussion， questioning， and answering questions， while ignoring the emotional expression， interesting presentation， scenario reproduction and talent show in the interactive process， making the live classroom lack of certain attraction and appeal. Based on this， the article makes an in-depth analysis of the status of classroom interaction in the current live classroom platform and points out the deficiencies in it. Then proposes a multi-element interactive optimization strategy for live classrooms， aiming at emotions， fun， games， situations， talents， etc. to enriches the interactive form of the live classroom and enhances the interactive effect of the live classroom， so that the teacher-student friendship can be enhanced， the teaching effect can be improved， and the student learning can be promoted.关键词：live classroom;multi-interactive;online learning;web-based teaching
- 8
- |
- 2
- |
- 0
发布时间：2023-05-05 - 摘要：Under the background of increasingly updated digital tools and resources， the effective design and realization of learning system is the key to improve learning services quality. However， the traditional requirement analysis methods such as questionnaire and interview hindered the user comprehensive， objective， real requirement. Therefore， introduce design thinking concept in the overall conception and design behavior to guide learning system design， at the same time make full use of real-time user online behavior data for user’s real "idea" and potential requirement， promote the efficiency of system design. Based on the research status of learning system design， this paper deconstructed the design thinking core stage， and discussed the mapping relationship about the behavior data， learner’s information and system measurement index， then guided by design thinking， based on education data， learning system as output results， constructed a frame for data driven learning system design by design thinking guidance， and around the "SQL learning system" to carry out case practice. The research showed that the iterative system has been significantly improved in terms of software efficiency and user interface， data-driven learning system design framework by design thinking guidance plays a significant role in obtaining the overall needs of users and the effective design and development of the system.关键词：design thinking;learning system;data driven;system design
- 10
- |
- 4
- |
- 0
发布时间：2023-05-05 - 摘要：Operating system is the core course of computer science major. However， for some colleges， due to the limitation of the number of teachers， it is difficult for teachers to provide one-to-one and targeted guidance for each student's homework and experiment. As a result， the teaching effect is difficult to achieve the expectation. In view of the above problems， discuss the teaching method of using cloud class to manage students' homework and using online evaluation platform to manage students' experiments in the teaching practice of operating system course， in order to improve the teaching quality. The online judge platform can automatically evaluate students' experiments and automatically generate experiment reports. Furthermore， teachers can design experiments reasonably， so that the online evaluation platform can automatically locate the types of errors in the program， so as to help students to complete the experiment independently and improve their practical ability. Teaching practice shows that those teaching methods can effectively improve the learning quality of students and has good application values.关键词：operating system;teaching practice;cloud class;online evaluation
- 13
- |
- 4
- |
- 0
发布时间：2023-05-05 - 摘要：During the cultivation of master graduate of discipline majoring in software engineering in local colleges and universities，the problems such as weak practical and innovation ability， poor social adaptability， unbalanced cultivation of students' ability， and difficult to meet the needs of enterprises and society exist. Therefore， firstly， define the elements of practical and innovation ability of master graduate of discipline majoring in software engineering， then puts forward the training process of double cycle practical and innovation ability， and finally focuses on the closed-loop training of professional practical courses. Through the practical application of the 2015-2019 students， the data analysis of the students' published papers， approved invention patents and software copyrights， and competition awards during their studies， it can be found that the students' practical innovation ability has been significantly improved； by analyzing the postgraduate enrollment data from 2015-2021 and the employment data of graduates from 2018-2021， it can be found that the degree site and the school's social influence are increasing year by year， and the school running strength of the degree site is improving year by year.关键词：master graduate of professional degree;software engineering;practical and innovation ability;personnel training
- 4
- |
- 3
- |
- 0
发布时间：2023-05-05 - 摘要：The future war is an operation under the condition of informatization， and informatization is the greatest actual combat. Based on the analysis of the internal unity of informatization and actual combat from the three aspects of combat environment， equipment and style， this paper combed the basic and key role of "University Computer Foundation" in the construction of curriculum system， the construction of new majors and the cultivation of military talents. The curriculum reform of "University Computer Foundation" under the background of practical combat， through the construction of a three-dimensional target system of "knowledge， ability， thinking and emotion"， strengthened the expansion and extension of teaching content from two aspects： the horizontal expansion of disciplines and majors and the vertical extension of post ability， and strengthen the application of information technology and open discussion and exchange in teaching methods. Through the two-phase pilot， students' military quality， information ability and leadership ability had been improved， and they have initially possessed the awareness of joint operations and computing thinking. The practical teaching has achieved good results， providing reference and guidance for the practical teaching reform of courses in military colleges and universities.关键词：practical teaching;information quality;university computer foundation;military academy
- 12
- |
- 4
- |
- 0
发布时间：2023-05-05 - 摘要：With the rapid development of Internet， computer and storage technology， the daily output of digital image information is exploding. Image retrieval is a hot research direction in the field of computer vision， which aims to retrieve and query data vision or text related content from large-scale image databases. Therefore， how to quickly and accurately retrieve user required content from massive digital information databases is an urgent problem in the field of image retrieval. The traditional low-level and depth features of images are the feature representations that effectively describe images. In recent years， the extraction of image depth features has attracted extensive attention and has developed rapidly in image retrieval technology in the field of computer vision. Therefore， by combining the advantages and disadvantages of TBIR and CBIR methods， the related research of CBIR technology in recent years is reviewed. Firstly， CBIR tasks and evaluation methods are introduced， and various classic datasets currently used in image retrieval tasks are summarized. Then， according to the image feature extraction methods， the relevant algorithms are introduced from the aspects of traditional and depth features， including the global and local feature extraction of the image and the feature extraction method based on the depth network model. Next， eight types of retrieval techniques， including cross modal， category level and instance level， are summarized. Finally， we summarize the problems that need to be solved in the current image retrieval technology， and analyze the future research direction of this technology on this basis.关键词：image retrieval;CBIR;deep learning;feature extraction;convolutional neural network
- 23
- |
- 4
- |
- 0
发布时间：2023-05-05 - 摘要：As a new intelligent bionic optimization algorithm， the flower pollination algorithm （FPA） has attracted extensive attention of researchers in the field of optimization algorithms and successfully applied to many practical problems in engineering fields due to its simple structure， fewer control parameters， and remarkable optimization performance. With constantly study， many improved FPA are proposed， mainly improve the search strategy and parameter control of the algorithm. Firstly， the structure of traditional FPA is described in detail， and the existing improved FPAs are summarized from the following three aspects： improved search strategy， integration of other ideas， and design parameter control. Secondly， comparative experiments are designed to compare the performance of several representative FPAs， which analyze the advantages and disadvantages of each algorithm in unimodal functions， multimodal functions， and complex functions， so as to make reference suggestions for engineers to choose algorithms when solving different practical problems. Finally， the problems that need to be further explored in the algorithm are pointed out.关键词：flower pollination algorithm;search strategy;parameter control;research overview
- 13
- |
- 3
- |
- 0
发布时间：2023-05-05
- 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