Top Read Articles

    Published in last 1 year |  In last 2 years |  In last 3 years |  All
    Please wait a minute...
    For Selected: Toggle Thumbnails
    Data Augmentation for Chinese Named Entity Recognition Task
    LI Jian, ZHANG Ke-liang, TANG Liang, XIA Rong-jing, REN Jing-jing
    Computer and Modernization    2022, 0 (04): 1-6.  
    Abstract368)            Save
    In low-resource natural language processing (NLP) tasks, the existing data is not enough to train an ideal deep learning model. Text data augmentation is an effective method to improve the training effect of such tasks. This paper proposes a group of data augmentation methods based on instance substitution for the task of Chinese named entity recognition. A named entity in the training sample can be replaced by another entity of the same kind without changing the label. The specific algorithms include: 1) crossover substitution between existing entities; 2) synonymous replacement of entity components; 3) automatic generation of Chinese names. These methods are applied to PeopleDailyNER and CLUENER2020 datasets respectively, and the augmentation data is used to train the BERT+CRF model. The experimental results show that the F1 value of the model can be improved by about 10% and 7% respectively on the two datasets with only adding the same amount of augmentation data as the original data under the condition of small samples, and it also has a significant improvement when the training samples increase.
    Reference | Related Articles | Metrics
    A Text Classification Model Based on BERT and Pooling Operation
    ZHNAG Jun, QIU Long-long
    Computer and Modernization    2022, 0 (06): 1-7.  
    Abstract359)            Save
    The fine-tuning method using the pre-trained language model has achieved good results in many natural language processing tasks represented by text classification, BERT model based on the Transformer framework as a typical representative especially. However, BERT uses the vector corresponding to [CLS] as the text representation directly, and does not consider the local features and global features of texts, which limits the classification performance of the model. Therefore, this paper proposes a text classification model that introduces a pooling operation, and uses pooling methods such as average pooling, maximum pooling, and K-MaxPooling to extract the representation vector of texts from the output matrix of BERT. The experimental results show that compared with the original BERT model, the text classification model with pooling operation proposed in this paper has better performance. In all text classification tasks in the experiment, its accuracy and F1-Score value are better than BERT model.
    Reference | Related Articles | Metrics
    An Improved Whale Optimization Algorithm Base on Hybrid Strategy
    LI Ru, FAN Bing-bing
    Computer and Modernization    2022, 0 (06): 13-20.  
    Abstract259)            Save
    In order to solve the problems of the original whale optimization algorithm (WOA) with slow convergence speed, weak global search ability, low solution accuracy and easy to fall into local optimization, a hybrid strategy is proposed to improve the whale optimization algorithm (LGWOA). Firstly, the Levy flight strategy is introduced into the position update formula of the whale random search, and the global search step is increased through Levy flight, the search space is enlarged, and the global search capability is improved. Secondly, the adaptive weight is introduced into the whale spiral upward position update formula to improve the algorithm’s local search ability and optimization accuracy. Finally, the idea combining the genetic algorithm’s cross mutation is used to balance the algorithm’s global search and local search capabilities, maintain the diversity of the population, and avoid falling into the local optimum. Simulation experiments on 12 benchmark test functions in different dimensions show that the improved whale algorithm has faster convergence speed and higher optimization accuracy.

    Reference | Related Articles | Metrics
    Categorical Data Clustering Based on Extraction of Associations from Co-association Matrix
    GUAN Yun-peng, LIU Yu-long
    Computer and Modernization    2022, 0 (11): 1-8.  
    Abstract253)            Save
    Categorical data clustering is widely used in different fields in the real world, such as medical science, computer science,  etc. The usual categorical data clustering is studied based on the dissimilarity measure. For data sets with different characteristics, the clustering results will be affected by the characteristics of the data set itself and noise information. In addition, the categorical data clustering based on representation learning is too complicated to implement, and the clustering results are greatly affected by the representation results. Based on the co-association matrix, this paper proposes a clustering method that can directly consider the relationship between the original information of categorical data, categorical data clustering based on extraction of associations from co-association matrix (CDCBCM). The co-association matrix can be regarded as a summary of the information association in the original data space. The co-association matrix is constructed by calculating the co-association frequency value of different objects in each attribute subspace, and some noise information is removed from the co-association matrix, and then the clustering result is obtained by normalized cut. The method is tested on 16 publicly available datasets in various aspects, compared with 8 existing methods, and detected using the F1-score metric. The experimental results show that this method has the best effect on 7 data sets, the average ranking is the best, and it can better complete the clustering task of categorical data.
    Reference | Related Articles | Metrics
    Character Network Analysis of Ordinary World
    WANG Jun, HE Jin-rong, MA Le-rong
    Computer and Modernization    2022, 0 (06): 32-36.  
    Abstract218)            Save
    The construction and quantitative analysis of the relationship network of characters in literary works is an important content of intelligent interpretation of literary works. This article takes Mr. Lu Yao’s literary work "The Ordinary World" as the research object and uses complex network analysis methods to construct and analyze the social network in the literary works. Firstly, the social network relationship in the work is extracted, where the characters in the novel correspond to the nodes in the network, the relationships between the characters correspond to the edges of the network, and the number of times the characters appear together in each chapter corresponds to the weight of the edges. Then we analyze the betweenness and aggregation coefficient correlation, hierarchical clustering and predicting link on the constructed network. The experimental results show that the character relationship network of Ordinary World is a heterogenous network with small-world characteristics. This research is helpful to promote the analysis of character relationship network in literary works.
    Reference | Related Articles | Metrics
    Dynamic Allocation Algorithm of Container Cloud Resources Based on Bi-level Programming
    ZHOU Yong-fu, XU Sheng-chao
    Computer and Modernization    2022, 0 (12): 1-5.  
    Abstract214)            Save
    The dynamic configuration decision problem of container cloud resources is analyzed in this paper. By defining the scheduling task of container cloud resources, the scheduling time of container source resources is solved. The shortest time matrix of container cloud resource scheduling task is used to obtain the conditions needed for container cloud resource scheduling. Under the bi-level planning condition, the objective function and constraint function of container cloud resource scheduling are solved, and the container cloud resource scheduling model is constructed. Considering the tasks of users and the cloud resources of data centers, a matrix to physical hosts is constructed on virtual machines. By constructing the objective function of container cloud resource dynamic configuration results in optimization, and combining with constraints, the dynamic configuration of container cloud resources is realized. Experimental results show that the proposed algorithm can not only improve the utilization of container cloud resources, but also reduce the configuration completion time, and has better dynamic configuration performance.
    Reference | Related Articles | Metrics
    Survey of Fruit Object Detection Algorithms in Computer Vision
    LI Wei-qiang, WANG Dong, NING Zheng-tong, LU Ming-liang, QIN Peng-fei
    Computer and Modernization    2022, 0 (06): 87-95.  
    Abstract207)            Save
    Fruit target detection and recognition based on computer vision is an important cross-disciplinary research topic of target detection, computer vision, agricultural robots, etc. It has important theoretical research significance and practical application value in the fields of smart agriculture, agricultural modernization, and automatic picking robots. As deep learning is widely used in the field of image processing and has achieved good results, fruit target detection and recognition algorithms combining computer vision technology with deep learning methods gradually become the mainstream. This article introduces the tasks, difficulties and development status of fruit target detection and recognition based on computer vision, as well as two types of fruit target detection and recognition algorithms based on deep learning methods. Finally, the public data set used for the training and learning of the algorithm model and the evaluation index for evaluating the performance of the model are introduced, and the current problems in the detection and recognition of fruit targets and the possible future development directions are discussed.
    Reference | Related Articles | Metrics
    PSWGAN-GP: Improved Wasserstein Generative Adversarial Network with Gradient Penalty
    CHEN Yun-xiang, WANG Wei, NING Juan, CHEN Yi-dan, ZHAO Yong-xin, ZHOU Qing-hua
    Computer and Modernization    2022, 0 (04): 21-26.  
    Abstract204)            Save
    The emergence of generative adversarial network (GAN) plays a great role in solving the problem of insufficient sample data in the field of deep learning. In order to solve the detail quality problems of images generated by GAN such as foreground and background separation and contour blurring, this paper proposes an improved Wasserstein generative adversarial network with gradient penalty (PSWGAN-GP) method. Based on the Wasserstein distance loss and gradient penalty of WGAN-GP, this method uses the features extracted from the three pooling layers of the VGG-16 network in the discriminator and calculates the style-loss and perceptual-loss from these features as penalty terms of the original loss, which improves the discriminator’s ability to acquire and discriminate deep features and enhance the details of the generated images. The experimental results show that PSWGAN-GP can effectively improve the quality of generated images with the same generator and discriminator network structure and the same hyperparameters, and the scores in IS and FID are improved relative to other image generation methods.
    Reference | Related Articles | Metrics
    Survey of Model Pruning Algorithms
    LI Yi, WEI Jian-guo, LIU Guan-wei
    Computer and Modernization    2022, 0 (09): 51-59.  
    Abstract190)            Save
    The model pruning algorithms apply different standards or methods to prune the redundant neurons in the deep neural network, which can compress the model to the maximum extent without losing the accuracy of the model, so as to reduce the storage and improve the speed. Firstly, the research status of model pruning algorithm and the main research direction are summarized and classified. The main research areas of model pruning include the granularity of pruning, the method to evaluate the importance of pruning elements, the sparsity of pruning, the theoretical foundation of model pruning, pruning for different tasks and so on. Then, the recent representative pruning algorithms are described in detail as well. Finally, the future research direction in this field is brought forward.
    Reference | Related Articles | Metrics
    Landslide Image Detection Based on Dilated Convolution and Attention Mechanism
    LIU Xue-hu, OU Ou, ZHANG Wei-jing, DU Xue-lei
    Computer and Modernization    2022, 0 (04): 45-51.  
    Abstract187)            Save
    Landslide area image detection and recognition has rich application and research value in disaster scope recognition, disaster data analysis and disaster prevention and mitigation. In this paper, a target detection method combining attention mechanism CBAM and dilated convolution is proposed to solve the problems of the diversity of landslide body shape and texture in landslide image and the unsatisfactory detection and recognition effect of landslide target area. On the basis of the traditional target detection algorithm Faster R-CNN, the attention mechanism model is added to the convolutional neural network layer. The landslide image features are extracted through the CBAM model combining spatial attention and channel attention, and the dilated convolution module is added to enlarge the receptive field area, and to improve the learning ability of the landslide target recognition and non-standard size in the remote sensing image area of the neural network, so as to further improve the detection accuracy of the landslide target area. The experimental results show that, based on the traditional target detection algorithm, the combination of the two methods can improve the recall rate and precision rate of target detection on the remote sensing images of landslides, and it has a certain validity and robustness.

    Reference | Related Articles | Metrics
    Helmet-wearing Detection Based on Improved YOLOv5
    YUE Heng, HUANG Xiao-ming, LIN Ming-hui, GAO Ming, LI Yang, CHEN Ling
    Computer and Modernization    2022, 0 (06): 104-108.  
    Abstract179)            Save
    To the problem that YOLOv5 cannot be focused by weights and cannot produce more distinguishable features, thereby reducing the accuracy of helmet detection, attention module was used. Besides, squeeze and excitation layer and efficient channel attention module were studied. To the problem that the non maximum suppression used by YOLOv5 to remove redundant results will only retain the highest confidence prediction frame of the same class when objects were highly overlapped, the Soft-NMS algorithm was used to keep more prediction boxes. Weighted non maximum suppression was used to fuse multiple prediction boxes information to improve the accuracy of the prediction boxes. For the problem of information loss caused by down-sampling , focus modules was used to improve the detection effect, and the various modules were integrated to obtain the optimal FESW-YOLO algorithm. Compared with YOLOv5, the algorithm improves the mAP@0.5 by 2.1 percentage points and the mAP@0.5:0.95 by1.2 percentage points on the helmet data set respectively, which improves the accuracy of safety helmet supervision.
    Reference | Related Articles | Metrics
    3D Dynamic Visualization Design and Simulation of Wind Field on Web
    TIAN Mao-chun, ZOU Xian-yong, YANG Yue, FAN Guang-wei, LAI Hang
    Computer and Modernization    2022, 0 (04): 97-102.  
    Abstract176)            Save
    In view of the difficulty in displaying the wind field quickly and visually on the Web, a Web-based 3D dynamic wind field visualization design and simulation method was proposed. Firstly, in order to solve the problem that wind field data is difficult to obtain and inconvenient to deal with, the process of wind field data acquisition, conversion, and application was given. Secondly, a method was designed to use Web Worker multi-threading technology to calculate and generate streamlines in parallel to reduce the generation time of wind field streamline tracing and improve the efficiency of streamline generation. Thirdly, combined with color mapping technology, a method to dynamically modify the streamline’s Alpha color channel to characterize the wind field movement was designed. Finally, the WebGL-based visualization engine Cesium was extended to perform the three-dimensional visualization rendering of the wind field, which can directly demonstrate the wind field. The results show that this method not only improves the efficiency of wind field visualization simulation on the Web, but also helps mitigating windstorm disaster.
    Reference | Related Articles | Metrics
    Text Clustering Algorithm Based on RoBERTa-WWM and HDBSCAN
    LIU kun, ZENG Xi, QIU Zi-heng, CHEN Zhou-guo,
    Computer and Modernization    2022, 0 (03): 48-52.  
    Abstract174)            Save
    In the big data environment, obtaining hot topics from massive Internet data is the basis for studying public opinion and sentiments in the current Internet. Among them, text clustering is one of the most common methods to get hot topics, which can be divided into two steps: text vectorization representation and clustering. However, in the task of vectorized text representation, the traditional text representation model cannot accurately represent the contextual information of texts such as news and posts. In the clustering task, the K-Means algorithm and DBSCAN algorithm are most commonly used, but their clustering method is not consistent with the actual distribution of topic data, which makes the existing text clustering algorithms very poorly applied in the actual Internet environment. Therefore, this paper proposes a text clustering algorithm based on RoBERTa-WWM and HDBSCAN according to the data distribution of topics in the Internet. Firstly, the pre-trained language model RoBERTa-WWM is used to obtain the text vector of each text. Secondly, the t-SNE algorithm is used to reduce the dimension of the high-dimensional text vector. Finally, the HDBSCAN algorithm based on hierarchical density clustering algorithm is used to cluster the low-dimensional text vector. The experimental results show that compared with the existing text clustering algorithms, the proposed algorithm has a great improvement in the clustering effect on data sets that contain noisy data and are unevenly distributed.
    Reference | Related Articles | Metrics
    Vehicle Target Detection Algorithm Based on YOLO v4
    YIN Yuan-qi, XU Yuan, XING Yuan-xin
    Computer and Modernization    2022, 0 (07): 8-14.  
    Abstract171)            Save
    Aiming at the problems of low occlusion target detection accuracy and poor small target detection effect in vehicle target detection, an improved target detection algorithm YOLO v4-ASC based on YOLO v4 is proposed.  By adding convolution block attention module to the tail of the backbone extraction network, the feature expression ability of the network model is improved; The loss function is improved to improve the convergence speed of the network model, and the Adam+SGDM optimization method is used to replace the original model optimization method SGDM to further improve the model detection performance. In addition, K-Means clustering algorithm is used to optimize the priori frame size, and the car, truck and bus categories in the traffic scene data set are combined as vehicle, which simplifies the problem in this paper into a two classification problem. The experimental results show that on the basis of maintaining the detection speed of the original algorithm, the proposed YOLO v4-ASC target detection algorithm achieves 70.05% AP and 71% F1-score. Compared with the original YOLO v4 algorithm, AP is improved by 9.92 percentage points  and F1 score is improved by 9 percentage points.
    Reference | Related Articles | Metrics
    Malicious TLS Traffic Identification Based on Deep Generation Adversarial Network
    QIN Ming-yue, NIAN Mei, ZHANG Jun,
    Computer and Modernization    2022, 0 (04): 121-126.  
    Abstract170)            Save
    The class imbalance problem in the public data sets of malicious encrypted traffic identification seriously affects the performance of malicious traffic prediction. In this paper, we propose to use the generator and discriminator in the depth generation adversarial network DGAN to simulate the generation of real data sets and the expansion of small sample data to form balanced data sets. In addition, in order to solve the problems that traditional machine learning methods rely on artificial feature extraction, which leads to the decrease of classification accuracy, a malicious traffic recognition model based on the combination of two-way gating loop unit BiGRU and attention mechanism is proposed. The deep learning algorithm automatically obtains the important feature vectors of different time series of data sets to identify malicious traffic. Experiments show that compared with the common malicious traffic recognition algorithms, the model has a good improvement in accuracy, recall, F1 and other indicators, and can effectively realize the identification of malicious encrypted traffic.
    Reference | Related Articles | Metrics
    Application of Hybrid CTC/Attention Model in Mandarin Recognition
    XU Hong-kui, ZHANG Zi-feng, LU Jiang-kun, ZHOU Jun-jie, HU Wen-ye, JIANG Tong-tong
    Computer and Modernization    2022, 0 (08): 1-6.  
    Abstract167)            Save
    The end-to-end speech recognition model based on Connectionist Temporal Classification (CTC) has the advantages of simple structure and automatic alignment, but the recognition accuracy needs to be further improved. This paper introduces the attention mechanism to form a hybrid CTC/Attention end-to-end model. This method adopts the multi-task learning approach, combining the alignment advantage of CTC with the context modeling advantage of attention mechanism. The experimental results show that when the 80-dimensional FBank feature and the 3-dimensional pitch feature are selected as the acoustic features, and the VGG-Bidirectional long short-time memory network is selected as the encoder for Chinese Mandarin recognition, the character error rate of this hybrid model is reduced by about 6.1% compared with the end-to-end model based on CTC, after the external language model is connected, the character error rate is further reduced by 0.3%. Compared with the traditional baseline model, the character error rate also decreased significantly.
    Reference | Related Articles | Metrics
    Privacy Protection Model for Medical Data Based on HISPAC
    YAO Zheng
    Computer and Modernization    2022, 0 (09): 1-12.  
    Abstract165)            Save
    The current era is computer time, especially the era of artificial intelligence and big data. The emergence of related industries has led to changes in various industries. As major service industry in China, the medical industry is changing quietly. At the some time the protection technology of medical privacy is developed continuously. With the explosrve growth of data, various types of patient identity information, case information and medical diagnosis information are leaked endless. In order to solve the topic of medical privacy protection, the paper constructs a set of medical privacy protection model, which includes two parts:1) An adaptive neural network privacy risk assessment model is constructed by using Recurrent Neural Network (RNN) and fuzzy reasoning theory, which is used to assign a credit label to the user’s behavior and calculate the privacy risk; 2) Based on the user credit risk value obtained from the model, a set of personal privacy data access control mechanism is established, namely HISPAC (Hospital Information System Privacy Access Control Model). The experiment proves that this mechanism has good privacy protection effect and can effectively solve the problem of medical data privacy leakage.
    Reference | Related Articles | Metrics
    Short-term Load Forecasting of Regional Microgrid Based on LSTM Neural Network
    YIN Chun-jie, XIAO Fa-da, LI Peng-fei, ZHAO Qin
    Computer and Modernization    2022, 0 (04): 7-11.  
    Abstract165)            Save
    There are many studies on the load forecasting of large power grids and relatively few studies on microgrids. Therefore, it is very important to establish a suitable microgrid load forecasting model to improve the accuracy of forecasting. This paper analyzes and selects temperature, daily type, and multiple historical loads as the input variables of the model for the case of fewer input variables, selects the LSTM neural network based on the recurrent neural network for modeling, and constructs the load forecasting model of microgrid based on LSTM neural network. Finally, in order to enhance the reliability of the results, two sets of load data in different time periods are used to predict separately, and the prediction results of the LSTM neural network are compared with those of BP neural network, RBF neural network and Elman neural network. The experimental results show that the prediction results of LSTM neural network are better than BP neural network, RBF neural network and Elman neural network. The LSTM neural network load forecasting model has good promotion prospect under the background of microgrid.

    Reference | Related Articles | Metrics
    FOCoR: A Course Recommendation Approach Based on Feature Selection Optimization
    WANG Yang, CHEN Mei, LI Hui
    Computer and Modernization    2022, 0 (10): 1-7.  
    Abstract162)            Save
    To solve the cold start problem of the recommendation model based on the behavioral log from online education platform, we design a course recommendation method named FOCoR that integrates data of course selection. First, we propose a technology of feature selection based on genetic algorithm (FSBGA), and then take the result of feature selection as input to build a recommendation model based on LightGBM which is a technology of gradient boosting tree for course recommendation. To be more specific, we construct a fitness function combining the loss of model and the number of features in the proposed FSBGA so that we successfully searched out the optimal feature subset that takes into account the loss of model and the number of features in the feature subset space of university course selection data. According to three indicators of log loss, F1-score and AUC, the model of course selection trained on the feature subset selected by the FSBGA is better than the models trained on the others selected by algorithms based on mutual information or F-test. In order to verify the effectiveness of the work in this paper, we have tested and evaluated FOCoR, LightGBM, XGBoost, decision tree, random forest, logistic regression and other algorithms on real data sets, and the results show that FOCoR has achieved the best performance in F1 scores.
    Reference | Related Articles | Metrics
    A Non-intrusive Load Monitoring Method Based on Improved kNN Algorithm and Transient Steady State Features
    TIAN Feng, DENG Xiao-ping, ZHANG Gui-qing, WANG Bao-yi
    Computer and Modernization    2022, 0 (10): 29-35.  
    Abstract156)            Save
    Non-intrusive load monitoring (NILM) can obtain the operation data of the electrical appliance in the circuit by analyzing the record from a single energy meter, which can serve as an important tool for energy saving planning and optimal dispatching for power grid. The existing NILM methods mainly focus on improving the accuracy of load identification, the model complexity is too high to be applied on embedded devices. A NILM method based on improved kNN algorithm and transient steady state feature is proposed to solve the above problems. Firstly, the kNN algorithm is selected as the load identification model because it does not require training, the kNN algorithm is improved by statistical method of distance weight, and the cosine similarity judgment mechanism is added to verify the accuracy of the kNN load identification results. Secondly, the transient and steady state features are selected as load characteristics to improve the identification of load features. Finally, experimental data are used to verify that the above NILM method has superior performance.
    Reference | Related Articles | Metrics
    Semi-supervised Learning Method Based on Convolution and Sparse Coding
    LIU Ying-jie, LAN Hai, WEI Xian
    Computer and Modernization    2022, 0 (11): 9-16.  
    Abstract155)            Save
    Convolutional neural network (CNN) has achieved great success in semi-supervised learning. It uses both labelled samples and unlabelled samples in the training stage. Unlabelled samples can help standardize the learning model. To further improve the feature extraction ability of semi-supervised models, this paper proposes an end-to-end semi-supervised learning method combining deep semi-supervised convolutional neural network and sparse coding dictionary learning, called Semi-supervised Learning based on Sparse Coding and Convolution (SSSConv), which aims to learn more discriminative image feature representation and improve the performance of classification tasks. Firstly, the proposed method uses CNN to extract features and performs orthogonal projection transformation on them. Then, learn the corresponding sparse coding and obtain the image representation. Finally, the classifier of the model can classify them. The whole semi-supervised learning process can be regarded as an end-to-end optimization problem. CNN part and sparse coding part have a unified loss function. In this paper, conjugate gradient descent algorithm, chain rule, and backpropagation algorithm are used to optimize the parameters of the objective function. Among them, we restrict the relevant parameters of sparse coding to the manifold, and the CNN parameters can be defined not only in Euclidean space but also in orthogonal space. Experimental results based on semi-supervised classification tasks verify the effectiveness of the proposed SSSConv framework, which is highly competitive with existing methods.
    Reference | Related Articles | Metrics
    High Illumination Visible Image Generation Based on Generative Adversarial Networks
    ZHUANG Wen-hua, TANG Xiao-gang, ZHANG Bin-quan, YUAN Guang-ming
    Computer and Modernization    2023, 0 (01): 1-6.  
    Abstract149)            Save
    To solve the problem of low accuracy of target detection under low illumination conditions at night, this paper proposes a generative adversarial network-based algorithm for high illumination visible light image generation. To improve the ability of the generator to extract features, a CBAM attention module is introduced in the converter module; To avoid the noise interference of artifacts in the generated images, the decoder of the generator is changed from the deconvolution method to the up-sampling method of nearest neighbour interpolation plus convolution layer; to improve the stability of the network training, the adversarial loss function is replaced from the cross-entropy function to the least-squares function. The generated visible images have the advantages of spectral information, rich detail information and good visibility enhancement compared with infrared images and night visible images, which can effectively obtain information about the target and scene. We verified the effectiveness of the method by image generation metrics and target detection metrics respectively, in which the mAP obtained from the test on the generated visible image improved by 11.7 percentage points and 30.2 percentage points respectively compared to the infrared image and the real visible image, which can effectively improve the detection accuracy and anti-interference capability of nighttime targets.
    Reference | Related Articles | Metrics
    Deep Recommendation Algorithm Integrating Triple Attention Mechanism and Review Score
    ZHANG Zong-hai, YU Yue-cheng, FENG Shen
    Computer and Modernization    2022, 0 (05): 1-9.  
    Abstract146)            Save
    Many shopping websites have a large amount of review information written by users. Although most recommendation systems use review information, there is still much room for improvement. On the one hand, the information in the comments is uneven, mixed with a lot of useless information; on the other hand, most existing recommendation systems assume that a user’s attention to a certain product feature is the same for all products and cannot accurately reflect user preferences. This paper proposes an aspect-aware depth recommendation model ANAP that integrates triple attention and review score. Starting from the two levels of words and features, the important information in the review text is extracted by constructing two different attention networks to reduce the impact of useless information; in order to accurately reflect user preferences, the attention interaction network is constructed to capture the user’s different attention to various aspects of different items, and to achieve fine-grained modeling of aspect perception. This paper conducts experiments on 6 real data sets and designs an attention mechanism comparison experiment. The results show that the ANAP model effectively improves the score prediction accuracy, and the mean absolute error (MAE) is lower than the existing best algorithm by 4.86 percentage points.
    Reference | Related Articles | Metrics
    Review on Fault Diagnosis Technology of Transformer
    LIN Fan-qin, LI Ming-ming, GUO Hong
    Computer and Modernization    2022, 0 (03): 116-126.  
    Abstract145)            Save
    Transformer equipment plays an essential role in the power system. Its healthy and stable operation is related to realizing the function of power distribution voltage conversion, and fault diagnosis technology can escort for the transformer’s regular operation. This paper summarizes the research status of transformer fault diagnosis technology at home and abroad, analyzes the development process of transformer fault diagnosis, compares the merits of different diagnosis methods, and analyzes the traditional method of extracting transformer fault data——dissolved gas extraction method and sound signal. Finally, the research focus and development trend of transformer fault diagnosis in the future are put forward to provide some reference for transformer fault diagnosis.
    Reference | Related Articles | Metrics
    Research Progress of Text Summarization Model
    ZHANG Zi-yun, WANG Wen-fa, MA Le-rong, DING Cang-feng
    Computer and Modernization    2022, 0 (06): 56-66.  
    Abstract141)            Save
    With more and more text data generated by the Internet, the problem of text information overload is becoming more and more serious. It is very necessary to reduce the dimension of various texts, and text summarization is one of the important means, and it is also one of the hot and difficult points in the field of artificial intelligence research. Text is designed to transform a text or a collection of texts into a short summary containing key information. In recent years, language model preprocessing has improved the technical level of many natural language processing tasks, including emotion analysis, question and answer, natural language reasoning, named entity recognition, text similarity and text summarization. In this paper, the classic methods of text summarization in the past and the methods of text summarization based on pre-training in recent years are combed, and the data sets and evaluation methods of text summarization are sorted out. Finally, the challenges and development trends of text summarization are summarized.
    Reference | Related Articles | Metrics
    Anti-collision Shortest Path Planning Based on Improved Dijkstra Algorithm
    HUANG Yi-hu, YU Ya-nan
    Computer and Modernization    2022, 0 (08): 20-24.  
    Abstract141)            Save
    Multiple UAVs may face the contradiction of track conflict when performing operational tasks. Therefore, an improved Dijkstra algorithm is proposed to realize the function of multiple UAVs to find the shortest and non-conflicting route. In the process of searching and traversing each track node by the classical Dijkstra algorithm, the variable length backtracking array of precursor nodes of each node is introduced to record all precursor nodes contained in each node, and all feasible shortest length routes from the starting point to the target point of each task are found. Then the time window conflict judgment model is introduced to separate the non-conflicting routes from all feasible routes of each task. Once all routes conflict, the conflict node in one of the shortest routes is treated as a temporary obstacle point and the shortest route that does not conflict with other tasks is re found by changing the backtracking array. Matlab software is used to design and write programs to verify the algorithm. The experiments show that the improved algorithm can plan all the shortest and non-conflicting routes contained in each task when multiple UAVs perform operational tasks and the planning efficiency of task set has been significantly improved.
    Reference | Related Articles | Metrics
    A Financial News Sentiment Analysis Method Based on Graph Convolutional Neural Network and Dependency Analysis
    YAO Chun-hua, ZHANG Xue-lei, SONG Xing-yu, ZHANG Ju, CAI Jia-zhi, FENG Ao
    Computer and Modernization    2022, 0 (05): 33-39.  
    Abstract140)            Save
    Sentiment analysis of financial news helps enterprises and investors to determine investment risks and improves economic benefits, resulting in high application value. Graph neural networks have excellent performance in text classification, and have been applied to the field of sentiment analysis. In this paper, we propose a sentiment analysis method that uses dependency syntax analysis in graph convolutional neural networks (Dependency Analysis-based Graph Convolutional Network, DA-GCN) for financial news. This method obtains the word order information of the sentence and the syntactic in the document by analyzing the dependency of words in the document. It then implements information propagation and weight updates in the graph with co-occurrence information in each document. Experiments on a financial news dataset show that our model achieves significant performance improvements over traditional deep learning methods.
    Reference | Related Articles | Metrics
    Social Bots Detection Based on Generative Adversarial Networks
    LI Yang-yang, YANG Ying-guang
    Computer and Modernization    2022, 0 (03): 1-6.  
    Abstract140)            Save
    Twitter is a social media with hundreds of millions of active users. Nearly 15% of bot accounts are controlled by automated programs. Some of these bot accounts are malicious account that spread malicious information. Although researchers have developed a large number of sophisticated bot account detection methods, they all require prior knowledge of bot accounts which are lack of generalization. In order to solve these problems, this paper proposes to use the discriminator from generative adversarial network for bot account detection. This makes it possible to obtain a good detection model with the examples of real accounts. Experiments on a popular dataset show that the AUC achieves 94% classification effect.
    Reference | Related Articles | Metrics
    ECG Signal Classification Based on Deep Learning
    YU Yan, QIU Lei,
    Computer and Modernization    2022, 0 (05): 16-20.  
    Abstract139)            Save
    Electrocardiogram (ECG) can reflect the state of the heart in real time, and can be used for the accurate diagnosis of arrhythmias and other cardiovascular diseases. In view of the noise interference during ECG signal acquisition, we reconstruct the fourth-order components of Db6 wavelet, then use Butterworth low pass filter to realize double denoising. Then, from denoised ECG signals to extract the R-wave, and the P-QRS-T are intercepted and input into the one-dimensional improved GoogLeNet model for training. One-dimensional improved GoogLeNet is an improved structure of the original two-dimensional GoogLeNet, which reduces the network depth and adds the maximum pooled layer and dilated convolution in the sparse connection to increase the receptive field, so as to reduce the amount of calculation and improve the training performance. Experiments on the MIT-BIH data set show that the classification accuracy is 99.39%, which is 0.17 percentage points and 0.22 percentage points higher than the one-dimensional GoogLeNet and the original GoogLeNet respectively, and the training efficiency is improved. Signal classification has a marked improvement over other advanced techniques.
    Reference | Related Articles | Metrics
    Multi-feature Fusion Fundus Image Segmentation Based on Codes Structure
    DING Wan-ying, CHEN Wei, LI Zhao-hui
    Computer and Modernization    2022, 0 (07): 1-7.  
    Abstract139)            Save
    In order to solve the problem that the existing fundus images segmentation methods have low segmentation precision and low accuracy for micro vessels, an improved U-Net network model based on codec structure is proposed. Firstly, the data is preprocessed and expanded, the green channel image is extracted, and the contrast is enhanced by contrasting limited histogram equalization and Gamma transform; Secondly, the training set is input into the neural network for segmentation, the residual module is added in the coding process, the high and low feature information are fused by short jump connection, the receptive field is increased by hole convolution, and the attention mechanism is added in the decoding module to increase the segmentation accuracy of fine blood vessels; Finally, the trained segmentation model is used to predict the retinal vascular segmentation results. Comparative experiments on DRIVE and CHASE-DB1 fundus image data sets show that the average accuracy, specificity and sensitivity of the model algorithm are 96.77% and 97.22%, 98.74% and 98.40%, 80.93% and 81.12% respectively. The results of experiments show that the algorithm can improve the accuracy and efficiency of microvascular segmentation, and can segment retinal vessels more accurately.
    Reference | Related Articles | Metrics
    Climate Change Prediction in Canada Based on VAR Model
    KOU Lu-yan, LIAO Jing, LI Xue-jun, WU Chang-shu, XIONG Jian-hua
    Computer and Modernization    2022, 0 (10): 13-18.  
    Abstract139)            Save
    The melting of Antarctic glaciers, the increasing of hurricanes and the gradual rising of sea level make people aware of the great challenges caused by global warming. So it is necessary to do research on global climate change. Missing data imputation is taken to study the data of four representative provinces in Canada, and a vector autoregressive (VAR) model is established considering the factors of solar radiation intensity, carbon dioxide content, soil water content, temperature, rainfall etc. to study Canada’s climate change. The specific model is established by doing stability test, impulse response and variance analysis, and is used to predict the temperature and precipitation in Canada. The experimental results show that the average temperature in Canada in the next 25 years will reach 15.0410 ℃, and the average precipitation will reach 2.0950 mm.
    Reference | Related Articles | Metrics
    A Point Cloud Registration Algorithm Combining Improved PSO Algorithm and TrICP Algorithm
    LIANG Zheng-you , , WANG Lu , LI Xuan-ang , YANG Feng ,
    Computer and Modernization    2022, 0 (05): 90-95.  
    Abstract137)            Save
    Aiming at the problem that the traditional iterative closest point (ICP) algorithm is easy to fall into the problem of local optimality when the initial spatial position deviation is large, a point cloud registration method combining improved PSO-TrICP algorithm is proposed. Firstly, the traditional particle swarm optimization (PSO) algorithm is improved by introducing similarity measurement criterion of fitness to adjust the updating mode of particles. Then, the mean value of the historical global optimal solution of each iteration is added as a new learning factor to avoid the phenomenon of “precocity”; Secondly, the rigid transformation parameters and the overlap rate between the point clouds are used to form the particles, and the improved PSO algorithm is used to provide a good initial relative position; Finally, the space transformation between point clouds is estimated with trimmed iterative closest point (TrICP) algorithm. Experimental results show that the improved PSO-TRICP algorithm has better registration accuracy and operation efficiency than the similar registration algorithms proposed in recent years, and has better robustness.
    Reference | Related Articles | Metrics
    Driver Distracted Behavior Recognition Based on Deep Learning
    HE Li-wen, ZHANG Rui-chi
    Computer and Modernization    2022, 0 (06): 67-74.  
    Abstract137)            Save
    Distracted driving behavior recognition is one of the main methods to improve driving safety. Aiming at the problem of low identification accuracy of distracted driving behavior, this paper proposes a driver distracted behavior recognition algorithm based on deep learning, which is composed of a cascade of target detection network and precise behavior recognition network. Based on the State Farm open data set, in the first level, the target detection algorithm SSD (Single Shot Multibox Detector) is used to extract local information from the original driver images in the data set and determine the candidate regions for behavior recognition. Then in the second level, the transfer learning VGG19, ResNet50 and MobileNetV2 models is used to accuratelyidentify the behavior information in the candidate region. Finally, the experiment compares the recognition accuracy of distracted driving behavior between layered recognition architecture and single model architecture. Results show that compared the proposed cascade network model with the mainstream model of single detection method, the driver behavior identification accuracy is improved 4% ~ 7% overall. Besides, the proposed algorithm not only reduces the influence of noise and other background regions on the model to improve the accuracy of distracted behavior recognition, but also can effectively identify more behavior categories to avoid the misclassification of actions.
    Reference | Related Articles | Metrics
    ERCUnet: An Improved Road Crack Detection Model Based on U-Net
    LIU Yu-xiang, SHE Wei, SHEN Zhan-feng, TAN Shuai
    Computer and Modernization    2022, 0 (07): 33-39.  
    Abstract135)            Save
    Aiming at the problems of traditional road  crack  detection methods,  such as low flexibility and poor universality, refering to the residual design in ResNet and the U-shaped encoding and decoding structure of U-Net model, an improved road crack detection model based on U-Net, named ERCUnet,  is designed. The model takes residual blocks as the main body, and optimizes the number of convolution cores of convolution layers at different depths for crack detection. All residual blocks in the model have the same structure, the overall structure of the model is more neat and simple, with good elasticity and strong structure. The residual structure not only makes the feature fusion more sufficient but also avoids the problem of gradient disappearance of deep convolution neural network. The experiment is conducted on the CrackForest dataset. The 118 labeled pictures of CrackForest are divided into training set and testing set according to the ratio of 5〖DK(〗∶〖DK)〗1. Through a series of data expansion methods, the problem of too little training data is effectively alleviated. The loss function combines cross entropy and F1 score to alleviate the imbalance between positive and negative samples. The final experimental results show that the number of parameters of ERCUnet model is only 13.30% of that of U-Net model, the recall, precision, and F1 are all greater than 70%, and noise rate and accuracy are 29.05% and 99.01% on testing set. ERCUnet-tiny model is obtained by modifying model parameters to confirm the plasticity of ERCUnet, and the number of its parameters is only 2.39% of that of U-Net model, similar effect to U-Net is achieved  on testing set.
    Reference | Related Articles | Metrics
    Network Intrusion Detection Model Based on Space-time Feature Fusion and Attention Mechanism
    RAO Hai-bing, ZHU Su-lei, YANG Chun-xia
    Computer and Modernization    2022, 0 (06): 116-121.  
    Abstract132)            Save
    Aiming at the problem of low network intrusion detection performance, a deep learning intrusion detection model CTA-net based on space-time feature fusion and attention mechanism is proposed. The model obtains space-time fusion features by integrating convolutional neural network (CNN) and long-short-term memory network (LSTM), and then uses the attention module (Attention) to calculate the importance of the input space-time fusion features, and finally passes the softmax function sort. Using the NSL-KDD data set, the experimental results show that compared with the CNN model with similar structure and the space-time fusion CNN-LSTM model, the convergence of the training set is significantly improved, and the accurate of classification evaluation index used on the test set  has increased by 10.9120 percentage points and 11.8740 percentage points, the precision has increased by 9.1950 percentage points  and 9.6130 percentage points, the recall has increased by 9.1780 percentage points  and9.9340 percentage points, and F1-SCORE has increased by 10.7830 percentage points  and 11.750 percentage points . The simulation results show that the proposed CTA-net model has good application potential in network intrusion detection.
    Reference | Related Articles | Metrics
    Microblog Rumor Detection Integrating User’s History and Dissemination Information
    LU Yue, CAO Chun-ping
    Computer and Modernization    2022, 0 (06): 37-42.  
    Abstract132)            Save
    With the development of Internet technology, online rumors have gradually spread on social media platforms based on Weibo. Research on the automatic detection of Weibo rumors is of great significance to maintaining social stability. The current mainstream rumor detection methods based on deep learning generally have the problem of not fully considering the semantic information of Weibo texts. At the same time, the rumor detection methods that rely too much on dissemination of information make the detection time lag and cannot meet the actual needs of rumor detection. In response to the above problems, this paper proposes a microblog rumor detection model that integrates user historical interaction information. It does not use the dissemination information of microblogs to be detected, constructs and trains the AbaNet (ALBERT-BiGRU-Attention) deep learning network model, and fully considers the text features and semantic information of Weibo and user history dissemination information text for rumor detection. The experimental results show that the model in this paper has the characteristics of high accuracy and strong stability, and can greatly shorten the time of rumor detection while obtaining high detection accuracy.
    Reference | Related Articles | Metrics
    Anomaly Detection of Student Consumption Data Based on Semi-supervised Learning
    SONG Xiao-li, ZHANG Yong-bo, ZHANG Pei-ying
    Computer and Modernization    2022, 0 (12): 13-17.  
    Abstract131)            Save
    With the more and more extensive application scenarios of campus card, the problem of capital security of campus card has become increasingly prominent. Campus card fraud will not only bring economic losses to teachers, students and businesses in the school, but also endanger the normal order of the campus. Aiming at the problem that the traditional anomaly detection method can not effectively extract the temporal feature of student consumption data, this paper proposes an anomaly detection method of student consumption data based on semi-supervised learning. Firstly, the auto-encoder is enhanced with the Gated Recurrent Unit, so that the model can reconstruct the consumption data more accurately. Then, the reconstruction error is calculated by Mahalanobis Distance, and the error threshold is determined by Fβ-Socre to detect abnormal data. Finally, the proposed method is used to detect the anomaly of student consumption data in a university. Experimental results show that the proposed method has better detection performance.
    Reference | Related Articles | Metrics
    On-line Detection System of CNC Tool Breakage Based on Attention-LSTM
    WANG Ke-yang, ZHANG Yao, LI Ke-wen, ZHANG Bao-qian, LI Jiang, REN Jie-wen
    Computer and Modernization    2022, 0 (05): 68-74.  
    Abstract129)            Save
    In order to detect tool damage during the batch processing of CNC machine tools to reduce defective products, an online monitoring method for tool damage in CNC production lines based on Attention-LSTM using machine tool spindle power information is proposed. The method uses the built-in sensor of the numerical control system as the data source to obtain the time series of the spindle power of the machine tool. In the data collection link, it is necessary to distinguish the different processes in the machining process and the tool number used in the process. Therefore, in the data acquisition link, the NC code and spindle power are collected at the same time, the collected data is processed by the NC code analysis method, the process identification of the processing process is completed, the Attention-LSTM algorithm is used to predict the spindle power data, and then the DTW algorithm is used to calculate the time series similarity. The degree of similarity between the processing power time series and the standard time series should be within a reasonable threshold range, otherwise it is considered that tool breakage occurred during the processing. Experiments were conducted on the FANUC CNC system to verify the accuracy of tool breakage recognition.
    Reference | Related Articles | Metrics
    Gait Feature Recognition Based on Improved Residual Network and Joint Loss Function
    HE Xuan, LIU Yi-xin, HE Xiao-hai, QING Lin-bo, CHEN Hong-gang
    Computer and Modernization    2022, 0 (04): 27-32.  
    Abstract126)            Save
    Aiming at the problems of insufficient recognition accuracy and shallow feature extraction level of the existing gait recognition models, a new joint loss gait feature recognition model Res-GaitSet based on improved residual network is proposed on the basis of GaitSet network. As a unique and effective biometric for long-distance recognition, gait can be widely used in geriatric evaluation, social order security and so on. In the new network, residual elements are introduced into the feature extraction module, and multiple loss functions are used together. This method effectively improves the accuracy and robustness of gait recognition model. The experimental results show that the accuracy of the improved network Res-GaitSet is improved in multiple scenes and different recognition angles of CASIA-B dataset. At the same time, the improved network is used for self built gait data set. Compared with the original network, the recognition effect of the improved network is also improved from different angles, which fully verifies the effectiveness of the improved model.
    Reference | Related Articles | Metrics
    Government Hotline Work-order Classification Fusing RoBERTa and Feature Extraction
    CHEN Gang
    Computer and Modernization    2022, 0 (06): 21-26.  
    Abstract125)            Save
    Government hotlines undertake a large number of citizens’ demands, which make manual work-order classification time-consuming and laborious. Most of the existing work-order classification methods are based on machine learning or single neural network model. With these methods, it is difficult to effectively understand the context semantic information, and the text feature extraction is not comprehensive. A government hotline work-order classification method fusing RoBERTa and feature extraction is proposed to address the above problems. The proposed method firstly obtains context-aware semantic feature vectors from textual descriptions of work-orders by RoBERTa pre-trained language model. Then, a feature extraction layer based on convolution neural network, bidirectional gated recurrent unit and Self-Attention mechanism is constructed to obtain the local and global features of the work-order semantic encodings, with the process of highlighting the semantic features with great importance for the global features. Finally, the fused feature vectors are input into the classifier to finish work-order classification. Experimental results show the proposed method can achieve better classification performance compared with several baseline methods.
    Reference | Related Articles | Metrics