Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906326
Weijun Wang, Haixia Pan, Yefan Cao
With the rapid development and popularization of mobile positioning devices such as mobile phones, location-based services play an increasingly important role in people’s production and life, which makes high-performance large-scale distance matrix calculation a key part of the optimization of many business scenarios. The distance matrix is used to calculate the shortest road distance of a group of starting and ending points in batches. Based on the Contraction Hierarchies algorithm, this paper proposes a method to quickly calculate the distance matrix between source nodes s and target nodes t, where the source nodes s $in$ S and target nodes t $in$ T. Under the scale of the road network in mainland China, take $|mathrm{S}|=|mathbf{T}|=10000$, the average solution time of the algorithm is only 7.2 minutes, which can meet the needs of various application scenarios.
{"title":"Large-Scale Distance Matrix Calculation Method Based on Contraction Hierarchies","authors":"Weijun Wang, Haixia Pan, Yefan Cao","doi":"10.1109/CCET55412.2022.9906326","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906326","url":null,"abstract":"With the rapid development and popularization of mobile positioning devices such as mobile phones, location-based services play an increasingly important role in people’s production and life, which makes high-performance large-scale distance matrix calculation a key part of the optimization of many business scenarios. The distance matrix is used to calculate the shortest road distance of a group of starting and ending points in batches. Based on the Contraction Hierarchies algorithm, this paper proposes a method to quickly calculate the distance matrix between source nodes s and target nodes t, where the source nodes s $in$ S and target nodes t $in$ T. Under the scale of the road network in mainland China, take $|mathrm{S}|=|mathbf{T}|=10000$, the average solution time of the algorithm is only 7.2 minutes, which can meet the needs of various application scenarios.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134090056","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906332
Ge Yuyao, Cheng Yiting, Wang Jia, zhou Hanlin, Chen Lizhe
In order to improve the ViT model accuracy of image classification task in Chinese medicine, this paper proposes a sharpening image preprocessing method of coupling residual algorithm, the image preprocessing method can make deep learning network makes it easier to extract the image edge character. In this paper, through a series of experiments to compare the algorithm under different parameters in traditional Chinese medicine classification accuracy of the data sets. Improved the vision Transformer structure of knowledge distillation and proposed the way of overlapping image blocks in PatchEmbeding operation to extract more information of the original image. A series of experiments were carried out on the traditional Chinese medicine data set. It is proved that the accuracy of the model is about 2% higher than that of the original knowledge distillation ViT structure.
{"title":"Vision Transformer Based on Knowledge Distillation in TCM Image Classification","authors":"Ge Yuyao, Cheng Yiting, Wang Jia, zhou Hanlin, Chen Lizhe","doi":"10.1109/CCET55412.2022.9906332","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906332","url":null,"abstract":"In order to improve the ViT model accuracy of image classification task in Chinese medicine, this paper proposes a sharpening image preprocessing method of coupling residual algorithm, the image preprocessing method can make deep learning network makes it easier to extract the image edge character. In this paper, through a series of experiments to compare the algorithm under different parameters in traditional Chinese medicine classification accuracy of the data sets. Improved the vision Transformer structure of knowledge distillation and proposed the way of overlapping image blocks in PatchEmbeding operation to extract more information of the original image. A series of experiments were carried out on the traditional Chinese medicine data set. It is proved that the accuracy of the model is about 2% higher than that of the original knowledge distillation ViT structure.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"165 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114870753","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906391
Sheng-Kai Lin
With the increase of the related facilities and services around the driver, the driver’s identity becomes more and more important, so the research on the driver’s identification is also increasing gradually. Since the emergence of the online car-hailing platform represented by Uber, people’s travel has become more and more convenient, but there have also been many problems surrounding the identity of the driver. For example, the actual information of the driver does not match the information registered on the platform, which increase safety risk for passengers. In this paper, we propose a novel driver identification scheme that first converts the raw data of the x and y axes of the accelerometer into feature vectors by a novel data transformation method which adds frequency domain perspective on the basis of time domain perspective, adopts sliding window and fast Fourier transform and then uses these feature vectors as the input of neural network. Finally, we identify the driver through our designed driver identification algorithm, which accepts as input the probability distribution of the network output. In our experiments, we experiment with 10 drivers and use accuracy, precision, and recall as outcome metrics. Experimental results show that the performance based on time and frequency features is better than that of time or frequency features alone.
{"title":"Driver Identification with Time and Frequency Features Derived from Vehicular Acceleration Data","authors":"Sheng-Kai Lin","doi":"10.1109/CCET55412.2022.9906391","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906391","url":null,"abstract":"With the increase of the related facilities and services around the driver, the driver’s identity becomes more and more important, so the research on the driver’s identification is also increasing gradually. Since the emergence of the online car-hailing platform represented by Uber, people’s travel has become more and more convenient, but there have also been many problems surrounding the identity of the driver. For example, the actual information of the driver does not match the information registered on the platform, which increase safety risk for passengers. In this paper, we propose a novel driver identification scheme that first converts the raw data of the x and y axes of the accelerometer into feature vectors by a novel data transformation method which adds frequency domain perspective on the basis of time domain perspective, adopts sliding window and fast Fourier transform and then uses these feature vectors as the input of neural network. Finally, we identify the driver through our designed driver identification algorithm, which accepts as input the probability distribution of the network output. In our experiments, we experiment with 10 drivers and use accuracy, precision, and recall as outcome metrics. Experimental results show that the performance based on time and frequency features is better than that of time or frequency features alone.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"85 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124132502","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906374
Kun Yu
Ontologies are crucial for data integration and information sharing. However, due to the different knowledge backgrounds of domain and ontology developers, the heterogeneity problem of multi-source ontology existence is more prominent, and ontology mapping is an important way to solve the ontology heterogeneity problem. However, the ontology similarity calculation methods among them still need to be improved in terms of accuracy or stability. In this paper, we propose an ontology similarity calculation method based on graph attention networks, which models ontologies as heterogeneous graph networks and uses the graph attention network model to introduce an attention mechanism to dynamically consider the influence of edge weights to achieve neighbor aggregation and perform similarity calculation. The experimental results show that this method has higher accuracy than the existing ontology similarity calculation methods.
{"title":"Similarity Computation of Heterogeneous Ontology Based on Graph Attention Network","authors":"Kun Yu","doi":"10.1109/CCET55412.2022.9906374","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906374","url":null,"abstract":"Ontologies are crucial for data integration and information sharing. However, due to the different knowledge backgrounds of domain and ontology developers, the heterogeneity problem of multi-source ontology existence is more prominent, and ontology mapping is an important way to solve the ontology heterogeneity problem. However, the ontology similarity calculation methods among them still need to be improved in terms of accuracy or stability. In this paper, we propose an ontology similarity calculation method based on graph attention networks, which models ontologies as heterogeneous graph networks and uses the graph attention network model to introduce an attention mechanism to dynamically consider the influence of edge weights to achieve neighbor aggregation and perform similarity calculation. The experimental results show that this method has higher accuracy than the existing ontology similarity calculation methods.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"69 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124223894","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906373
Xiao-qiang Wang, Chen Zhang, Xiao Chang
As smart cities continue to be developed, network traffic surges. Although edge computing improves the quality of service by shifting computing tasks from cloud computing centers to locations closer to the edge through a multi-layer distributed computing model, there are security risks in the face of central node failures and internal malicious attacks, which make it difficult to provide stable and reliable services. In this paper, we design a trusted management model of edge device in smart city with blockchain, introducing blockchain technology into the research of smart city construction. It uses the blockchain distributed architecture and decentralization idea to realize the trusted collection and storage of sensory data. And based on this architecture, a new reputation-based PoW consensus algorithm is proposed to provide a trust mechanism for IoT devices. The algorithm greatly increases the cost of malicious attacks launched by the IoT device, effectively prevents malicious attacks on devices, and achieves trusted management of edge device behavior. The smart city application based on the proposed method verifies its feasibility, effectively prevents malicious attacks from nodes, and enhances the information security of the system.
{"title":"Trusted Management Infrastructure with Blockchain for Edge Device in Smart City","authors":"Xiao-qiang Wang, Chen Zhang, Xiao Chang","doi":"10.1109/CCET55412.2022.9906373","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906373","url":null,"abstract":"As smart cities continue to be developed, network traffic surges. Although edge computing improves the quality of service by shifting computing tasks from cloud computing centers to locations closer to the edge through a multi-layer distributed computing model, there are security risks in the face of central node failures and internal malicious attacks, which make it difficult to provide stable and reliable services. In this paper, we design a trusted management model of edge device in smart city with blockchain, introducing blockchain technology into the research of smart city construction. It uses the blockchain distributed architecture and decentralization idea to realize the trusted collection and storage of sensory data. And based on this architecture, a new reputation-based PoW consensus algorithm is proposed to provide a trust mechanism for IoT devices. The algorithm greatly increases the cost of malicious attacks launched by the IoT device, effectively prevents malicious attacks on devices, and achieves trusted management of edge device behavior. The smart city application based on the proposed method verifies its feasibility, effectively prevents malicious attacks from nodes, and enhances the information security of the system.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124051915","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906335
Qingshu Li, Xiǎohóng Shí, Qi Xu, Wei Huang, Peng Yang
Feature Pyramid Network (FPN) is a basic but important component in target detection system. Together with target detection algorithms, such as SSD, Faster R-CNN and YOLO series, they have achieved good detection results for large targets with high resolutions, but the performance is less effective when it comes to detect small targets that contain relatively little semantic information. And small target detection is quite common in daily life, such as face recognition at long distances, traffic sign detection in automatic driving, etc. That means it is significant to break the bottleneck of target detection and get better accuracy performance. In this article, based on the FPN, we propose an improved network structure (IMFPN) from two aspects to get a better accuracy result in small target detection task. In the first aspect, we improve the feature map pyramid structure for feature enhancement, reduce the problem of information loss during feature map fusion and get the semantic information of multiscale feature maps. In the second aspect, we concentrate on the problem of information loss in the pooling process of feature maps, we propose an improved version of the PRRoI pooling method that combines RoI Pooling and RoI Align Pooling. And we also optimize the positioning of the frame through a new IoU calculation standard. Based on these above ideas and methods, we propose a small target detection method based on feature enhancement and positioning optimization.
{"title":"A Small Target Detection Method Based on Feature Enhancement and Positioning Optimization","authors":"Qingshu Li, Xiǎohóng Shí, Qi Xu, Wei Huang, Peng Yang","doi":"10.1109/CCET55412.2022.9906335","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906335","url":null,"abstract":"Feature Pyramid Network (FPN) is a basic but important component in target detection system. Together with target detection algorithms, such as SSD, Faster R-CNN and YOLO series, they have achieved good detection results for large targets with high resolutions, but the performance is less effective when it comes to detect small targets that contain relatively little semantic information. And small target detection is quite common in daily life, such as face recognition at long distances, traffic sign detection in automatic driving, etc. That means it is significant to break the bottleneck of target detection and get better accuracy performance. In this article, based on the FPN, we propose an improved network structure (IMFPN) from two aspects to get a better accuracy result in small target detection task. In the first aspect, we improve the feature map pyramid structure for feature enhancement, reduce the problem of information loss during feature map fusion and get the semantic information of multiscale feature maps. In the second aspect, we concentrate on the problem of information loss in the pooling process of feature maps, we propose an improved version of the PRRoI pooling method that combines RoI Pooling and RoI Align Pooling. And we also optimize the positioning of the frame through a new IoU calculation standard. Based on these above ideas and methods, we propose a small target detection method based on feature enhancement and positioning optimization.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"974 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116203658","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906362
Zhaoyuan Liu, Jingyi Fan, S. Geng, Peng Qin, Xiongwen Zhao
In mobile edge computing (MEC) network, computing offloading can alleviate resource constraints and improve service quality effectively. Meanwhile data transmission in Device-to-Device (D2D) communication can realize resource sharing and balance among different users, in order to improve system spectrum utilization and reduce communication delay. Aiming at minimizing the task delay and terminal energy consumption in MEC network, a joint optimization problem considering task offloading and computing resource allocation is formulated in D2D-assisted MEC network by building cost optimization model with multi-objective constraints. Moreover, the original problem is decoupled as two sub-problems which are solved by discrete ternary particle swarm optimization (DTPSO) algorithm and Lagrange multiplier method, respectively. Simulation results show that compared with three other typical methods, the proposed scheme in this work can effectively reduce task execution delay and terminal energy consumption.
{"title":"Joint optimization of Task Offloading and Computing Resource Allocation in MEC-D2D Network","authors":"Zhaoyuan Liu, Jingyi Fan, S. Geng, Peng Qin, Xiongwen Zhao","doi":"10.1109/CCET55412.2022.9906362","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906362","url":null,"abstract":"In mobile edge computing (MEC) network, computing offloading can alleviate resource constraints and improve service quality effectively. Meanwhile data transmission in Device-to-Device (D2D) communication can realize resource sharing and balance among different users, in order to improve system spectrum utilization and reduce communication delay. Aiming at minimizing the task delay and terminal energy consumption in MEC network, a joint optimization problem considering task offloading and computing resource allocation is formulated in D2D-assisted MEC network by building cost optimization model with multi-objective constraints. Moreover, the original problem is decoupled as two sub-problems which are solved by discrete ternary particle swarm optimization (DTPSO) algorithm and Lagrange multiplier method, respectively. Simulation results show that compared with three other typical methods, the proposed scheme in this work can effectively reduce task execution delay and terminal energy consumption.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126591449","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906325
Tong Wu, K. Song, Hongwei Zhao
Software clock data recovery (CDR) is a critical component of a high-speed serial link that recovers the reference clock from a serial signal, which are generally used in real-time sampling oscilloscopes and electronic design automation software. The common method for calculating the period of the reference clock is to extract the rising and falling edges of the signal, and then calculate the difference between adjacent transition edges, so that a pulse width sequence is obtained. By performing statistics on this sequence, it can be found that the value of the position of the first peak of the statistical histogram is the period of the reference clock, the position of the second peak is twice the period of the clock, etc. The period of the clock can be calculated by weighted average. However, high-speed serial signals inevitably have duty-cycle-distortion (DCD) jitter, and DCD seriously affect the determination of the position of the peak of the above statistical histogram, there are many spurious peaks near the actual peak. This article proposes an improved method of calculating the clock period, which can eliminate the influence of DCD and to obtain an accurate clock period.
{"title":"An Improved Clock Cycle Measurement Method for High-Speed Serial Signal with Duty-Cycle-Distortion Jitter","authors":"Tong Wu, K. Song, Hongwei Zhao","doi":"10.1109/CCET55412.2022.9906325","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906325","url":null,"abstract":"Software clock data recovery (CDR) is a critical component of a high-speed serial link that recovers the reference clock from a serial signal, which are generally used in real-time sampling oscilloscopes and electronic design automation software. The common method for calculating the period of the reference clock is to extract the rising and falling edges of the signal, and then calculate the difference between adjacent transition edges, so that a pulse width sequence is obtained. By performing statistics on this sequence, it can be found that the value of the position of the first peak of the statistical histogram is the period of the reference clock, the position of the second peak is twice the period of the clock, etc. The period of the clock can be calculated by weighted average. However, high-speed serial signals inevitably have duty-cycle-distortion (DCD) jitter, and DCD seriously affect the determination of the position of the peak of the above statistical histogram, there are many spurious peaks near the actual peak. This article proposes an improved method of calculating the clock period, which can eliminate the influence of DCD and to obtain an accurate clock period.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"122 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114166681","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906389
Zibo Yi, Qingbo Wu, Jie Yu, Yongtao Tang, Xiaodong Liu, Long Peng, Jun Ma
In recent years, with the development of Tibetan language information technologies, the Internet Tibetan data is increasing year by year. Due to the need for the Tibetan input method and Tibetan error correction, Tibetan language prediction has become an urgent problem to be solved. At present, the challenges of Tibetan prediction are that the Tibetan syllable composition is complex, the vocabulary of Tibetan words which is composed of syllables is extremely large, and the Tibetan word separation technology is not mature. To solve the above problems, this paper proposes a Tibetan syllable prediction method based on a pre-trained cross-lingual language model using Tibetan syllables instead of Tibetan words as the token for prediction. The method uses the cross-lingual language model XLM-R and fine-tunes it using Tibetan news texts to make it more suitable for predicting Tibetan in the news domain. We conduct experiments on Tibetan syllable prediction for texts crawled on the Tibetan news website. The experiments show that the precision of our model for Tibetan text prediction is higher than that of the current n-gram methods.
{"title":"Tibetan Syllable Prediction with Pre-trained Cross-lingual Language Model","authors":"Zibo Yi, Qingbo Wu, Jie Yu, Yongtao Tang, Xiaodong Liu, Long Peng, Jun Ma","doi":"10.1109/CCET55412.2022.9906389","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906389","url":null,"abstract":"In recent years, with the development of Tibetan language information technologies, the Internet Tibetan data is increasing year by year. Due to the need for the Tibetan input method and Tibetan error correction, Tibetan language prediction has become an urgent problem to be solved. At present, the challenges of Tibetan prediction are that the Tibetan syllable composition is complex, the vocabulary of Tibetan words which is composed of syllables is extremely large, and the Tibetan word separation technology is not mature. To solve the above problems, this paper proposes a Tibetan syllable prediction method based on a pre-trained cross-lingual language model using Tibetan syllables instead of Tibetan words as the token for prediction. The method uses the cross-lingual language model XLM-R and fine-tunes it using Tibetan news texts to make it more suitable for predicting Tibetan in the news domain. We conduct experiments on Tibetan syllable prediction for texts crawled on the Tibetan news website. The experiments show that the precision of our model for Tibetan text prediction is higher than that of the current n-gram methods.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129242287","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-19DOI: 10.1109/CCET55412.2022.9906393
Li Guo, Zhongyue Chen, Xiaoping Chen
With the development of semantic segmentation, segmentation-based methods have yielded great success in detecting arbitrary-shaped texts. However, many existing text detection methods use binary discrete distributions to predict shrunk text instances, which cannot generate complete and accurate text bounding boxes. In this paper, we propose an arbitrary-shaped scene text detection method based on predicting Gaussian probability distance map of the complete text region, and this map can retain more text boundary information. Then, the boundary pixels are clustered into high-confidence text centers by a learnable post-processing and false positives are filtered out by pixel-level score maps. We also propose an adaptive channel enhancement module to improve the pixel-level segmentation accuracy. Experiments on three standard datasets, including CTW1500, Total-Text, and MSRA-TD500, demonstrate that the proposed method achieves great robustness and performance. The method obtains an F-measure of S2.S% on CTW1500 and S3.0% on MSRA-TD500.
{"title":"Arbitrary-Shaped Text Detection with Gaussian Probability Distance Distribution","authors":"Li Guo, Zhongyue Chen, Xiaoping Chen","doi":"10.1109/CCET55412.2022.9906393","DOIUrl":"https://doi.org/10.1109/CCET55412.2022.9906393","url":null,"abstract":"With the development of semantic segmentation, segmentation-based methods have yielded great success in detecting arbitrary-shaped texts. However, many existing text detection methods use binary discrete distributions to predict shrunk text instances, which cannot generate complete and accurate text bounding boxes. In this paper, we propose an arbitrary-shaped scene text detection method based on predicting Gaussian probability distance map of the complete text region, and this map can retain more text boundary information. Then, the boundary pixels are clustered into high-confidence text centers by a learnable post-processing and false positives are filtered out by pixel-level score maps. We also propose an adaptive channel enhancement module to improve the pixel-level segmentation accuracy. Experiments on three standard datasets, including CTW1500, Total-Text, and MSRA-TD500, demonstrate that the proposed method achieves great robustness and performance. The method obtains an F-measure of S2.S% on CTW1500 and S3.0% on MSRA-TD500.","PeriodicalId":329327,"journal":{"name":"2022 IEEE 5th International Conference on Computer and Communication Engineering Technology (CCET)","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123518508","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：