Pub Date : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00030
Xihai Xie, Mingxin Zhang
The traditional non-coherent Multi-Symbol Detection (MSD) method for Continuous Phase Frequency Shift Keying (CPFSK) signals suffers from the problems of a large number of associated operations and high complexity. In this paper, we propose a non-coherent multi-symbol detection algorithm with reduced complexity. The improved algorithm selects only the local reference signal whose first symbol is the same as the already detected symbol for the correlation operation, which reduces the correlation operation by nearly 50% compared with the traditional algorithm and reduces the complexity of the algorithm. A graphical system modeling and simulation of the non-coherent multi-symbol detection algorithm with reduced complexity are performed in the DSP Builder platform. The model is simulated by using ModelSim software for engineering validation, and the simulation result shows that the model can implement the improved algorithm for CPFSK signal detection.
{"title":"Implementation of the CPFSK Signal Non-coherent Multi-symbol Detection Algorithm with Reduced Complexity","authors":"Xihai Xie, Mingxin Zhang","doi":"10.1109/ICNLP58431.2023.00030","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00030","url":null,"abstract":"The traditional non-coherent Multi-Symbol Detection (MSD) method for Continuous Phase Frequency Shift Keying (CPFSK) signals suffers from the problems of a large number of associated operations and high complexity. In this paper, we propose a non-coherent multi-symbol detection algorithm with reduced complexity. The improved algorithm selects only the local reference signal whose first symbol is the same as the already detected symbol for the correlation operation, which reduces the correlation operation by nearly 50% compared with the traditional algorithm and reduces the complexity of the algorithm. A graphical system modeling and simulation of the non-coherent multi-symbol detection algorithm with reduced complexity are performed in the DSP Builder platform. The model is simulated by using ModelSim software for engineering validation, and the simulation result shows that the model can implement the improved algorithm for CPFSK signal detection.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81729167","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00048
Xiaoming Lu, Wenjian Liu, Shengyi Jiang, Changqing Liu
Natural Language Processing (NLP) systems have three main components including tokenization, embedding, and model architectures (top deep learning models such as BERT, GPT-2, or GPT-3). In this paper, the authors attempt to explore and sum up possible ways of fine-tuning the Multilingual BERT (mBERT) model and feeding it with effective encodings of Tangut characters. Tangut is an extinct low-resource language. We expect to introduce a tailored embedding layer into Tangut as part of the fine-tuning procedure without altering mBERT internal structure. The initial work is listed on. By reviewing existing State of the Art (SOTA) approaches, we hope to further analyze the performance boost of mBERT when applied to low-resource languages.
{"title":"Multilingual BERT Cross-Lingual Transferability with Pre-trained Representations on Tangut: A Survey","authors":"Xiaoming Lu, Wenjian Liu, Shengyi Jiang, Changqing Liu","doi":"10.1109/ICNLP58431.2023.00048","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00048","url":null,"abstract":"Natural Language Processing (NLP) systems have three main components including tokenization, embedding, and model architectures (top deep learning models such as BERT, GPT-2, or GPT-3). In this paper, the authors attempt to explore and sum up possible ways of fine-tuning the Multilingual BERT (mBERT) model and feeding it with effective encodings of Tangut characters. Tangut is an extinct low-resource language. We expect to introduce a tailored embedding layer into Tangut as part of the fine-tuning procedure without altering mBERT internal structure. The initial work is listed on. By reviewing existing State of the Art (SOTA) approaches, we hope to further analyze the performance boost of mBERT when applied to low-resource languages.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81833326","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00033
Xi Hai Xie, Wen Chuan Wang
In order to improve the accuracy of speech recognition, the input speech signal is usually denoised first, which is typically done using the Least Mean Square (LMS) algorithm. To address the drawback that the fixed-step LMS algorithm in adaptive filtering cannot achieve a balance between convergence speed and steady-state error, this paper proposes a variable-step LMS algorithm based on an improved inverse hyperbolic sine function. In this paper, the improved algorithm is applied to speech enhancement, and the performance of this algorithm is compared with several other improved algorithms. The simulation results show that the improved algorithm takes better care of the conflict between convergence speed and steady-state error, and the algorithm has an obvious denoising effect for noisy speech, which effectively improves the clarity and intelligibility of speech and provides prerequisites for speech recognition.
{"title":"An Improved LMS Adaptive Filtering Speech Enhancement Algorithm","authors":"Xi Hai Xie, Wen Chuan Wang","doi":"10.1109/ICNLP58431.2023.00033","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00033","url":null,"abstract":"In order to improve the accuracy of speech recognition, the input speech signal is usually denoised first, which is typically done using the Least Mean Square (LMS) algorithm. To address the drawback that the fixed-step LMS algorithm in adaptive filtering cannot achieve a balance between convergence speed and steady-state error, this paper proposes a variable-step LMS algorithm based on an improved inverse hyperbolic sine function. In this paper, the improved algorithm is applied to speech enhancement, and the performance of this algorithm is compared with several other improved algorithms. The simulation results show that the improved algorithm takes better care of the conflict between convergence speed and steady-state error, and the algorithm has an obvious denoising effect for noisy speech, which effectively improves the clarity and intelligibility of speech and provides prerequisites for speech recognition.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81118513","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00064
Hongda Gong, Shimin Shan, Hongkui Wei
Graph-to-text generation task is transforms knowledge graphs into natural language. In current research, pretrained language models(PLMs) have shown better performance than structured graph encoders in the generation task. Currently, PLMs serialise knowledge graphs mostly by transforming them into undirected graph structures. The advantage of an undirected graph structure is that it provides a more comprehensive representation of the information in knowledge graph, but it is difficult to capture the dependencies between entities, so the information represented may not be accurate. Therefore, We use four types of positional embedding to obtain both the directed and undirected structure of the knowledge graph, so that we can more fully represent the information in knowledge graph, and the dependencies between entities. We then add a semantic aggregation module to the Transformer layer of PLMs, which is used to obtain a more comprehensive representation of the information in knowledge graph, as well as to capture the dependencies between entities. Thus, our approach combines the advantages of both directed and undirected structural information. In addition, our new approach is more capable of capturing generic knowledge and can show better results with small samples of data.
{"title":"Graph-to-Text Generation Combining Directed and Undirected Structural Information in Knowledge Graphs*","authors":"Hongda Gong, Shimin Shan, Hongkui Wei","doi":"10.1109/ICNLP58431.2023.00064","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00064","url":null,"abstract":"Graph-to-text generation task is transforms knowledge graphs into natural language. In current research, pretrained language models(PLMs) have shown better performance than structured graph encoders in the generation task. Currently, PLMs serialise knowledge graphs mostly by transforming them into undirected graph structures. The advantage of an undirected graph structure is that it provides a more comprehensive representation of the information in knowledge graph, but it is difficult to capture the dependencies between entities, so the information represented may not be accurate. Therefore, We use four types of positional embedding to obtain both the directed and undirected structure of the knowledge graph, so that we can more fully represent the information in knowledge graph, and the dependencies between entities. We then add a semantic aggregation module to the Transformer layer of PLMs, which is used to obtain a more comprehensive representation of the information in knowledge graph, as well as to capture the dependencies between entities. Thus, our approach combines the advantages of both directed and undirected structural information. In addition, our new approach is more capable of capturing generic knowledge and can show better results with small samples of data.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"86472407","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00053
Nurulhuda Mohamad Ali, Goh Hui Ngo, Amy Lim Hui Lan
Part-of-Speech (POS) Tagging is one of the fundamental tasks in Natural Language Processing (NLP) in analyzing human languages. It is a process of identifying how words are used in a sentence by assigning the proper POS for each word. Thus far, most well-researched POS tagging is on European languages which are considered rich-resource languages due to the unlimited linguistic resources such as research studies and large standard corpus. However, POS tagging is arduous for low-resource languages due to the limitation of linguistic resources. The Malay language is considered as a low-resource language. Most POS tagging studies for the Malay language are using rule-based and stochastic methods. However, exploration in Deep Learning (DL) for Malay language is limited. Thus, studies with POS tagging methods that implement DL for other low-resource languages within South East Asia are included in this study. Hence, the aim of this study is to identify the state of the art, challenges, and future works of Malay POS tagger. This study provides a review of different methods, datasets, and performance measures used in POS tagging studies.
{"title":"Construction of Part of Speech Tagger for Malay Language: A Review","authors":"Nurulhuda Mohamad Ali, Goh Hui Ngo, Amy Lim Hui Lan","doi":"10.1109/ICNLP58431.2023.00053","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00053","url":null,"abstract":"Part-of-Speech (POS) Tagging is one of the fundamental tasks in Natural Language Processing (NLP) in analyzing human languages. It is a process of identifying how words are used in a sentence by assigning the proper POS for each word. Thus far, most well-researched POS tagging is on European languages which are considered rich-resource languages due to the unlimited linguistic resources such as research studies and large standard corpus. However, POS tagging is arduous for low-resource languages due to the limitation of linguistic resources. The Malay language is considered as a low-resource language. Most POS tagging studies for the Malay language are using rule-based and stochastic methods. However, exploration in Deep Learning (DL) for Malay language is limited. Thus, studies with POS tagging methods that implement DL for other low-resource languages within South East Asia are included in this study. Hence, the aim of this study is to identify the state of the art, challenges, and future works of Malay POS tagger. This study provides a review of different methods, datasets, and performance measures used in POS tagging studies.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87004278","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00060
Sheping Zhai, Hang Li, Fangyi Li, Xinnian Kang
Extracting entities and relations from text is crucial in the field of natural language processing. Current methods for relation extraction rely on training sets labeled using remote supervision techniques. However, these methods have limitations as they do not consider the connection between entity and relation extraction and cannot extract overlapping entities and relations. Therefore, accurate joint entity-relation extraction remains challenging. Our paper introduces a model for entity relation extraction based on multi-head attention and graph convolutional networks. We utilize the multi-head attention approach to extract entity features, building on the text features extracted by the graph convolutional network. Utilizing the New York Times (NYT) dataset, we evaluated the performance of our model. The experimentation revealed that our model effectively captures the semantic correlation between entity and relation extraction and minimizes the impact of unrelated entity pairings, resulting in improved recognition accuracy even in scenarios with overlapping entities.
{"title":"Entity Relationship Extraction Method Based on Multi-head Attention and Graph Convolutional Network","authors":"Sheping Zhai, Hang Li, Fangyi Li, Xinnian Kang","doi":"10.1109/ICNLP58431.2023.00060","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00060","url":null,"abstract":"Extracting entities and relations from text is crucial in the field of natural language processing. Current methods for relation extraction rely on training sets labeled using remote supervision techniques. However, these methods have limitations as they do not consider the connection between entity and relation extraction and cannot extract overlapping entities and relations. Therefore, accurate joint entity-relation extraction remains challenging. Our paper introduces a model for entity relation extraction based on multi-head attention and graph convolutional networks. We utilize the multi-head attention approach to extract entity features, building on the text features extracted by the graph convolutional network. Utilizing the New York Times (NYT) dataset, we evaluated the performance of our model. The experimentation revealed that our model effectively captures the semantic correlation between entity and relation extraction and minimizes the impact of unrelated entity pairings, resulting in improved recognition accuracy even in scenarios with overlapping entities.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79205296","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00026
Chen Yang, Guang-Yuan Zhao
In order to solve the problem of low accuracy of license plate detection, an improved license plate detection algorithm is proposed. The super-resolution reconstruction network SRGAN is used to enhance the image of the dataset and make the image of the license plate area clearer; The fourth C3 module of YOLOv5s backbone network is replaced with CBAM attention mechanism module to enhance the ability of backbone network to extract feature information, thus improving the detection accuracy. The experimental results show that YOLOv5s network using SRGAN for image enhancement and embedding CBAM attention mechanism improves the accuracy of license plate image.
{"title":"Implementation of License Plate Detection Based on Improved YOLOv5s","authors":"Chen Yang, Guang-Yuan Zhao","doi":"10.1109/ICNLP58431.2023.00026","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00026","url":null,"abstract":"In order to solve the problem of low accuracy of license plate detection, an improved license plate detection algorithm is proposed. The super-resolution reconstruction network SRGAN is used to enhance the image of the dataset and make the image of the license plate area clearer; The fourth C3 module of YOLOv5s backbone network is replaced with CBAM attention mechanism module to enhance the ability of backbone network to extract feature information, thus improving the detection accuracy. The experimental results show that YOLOv5s network using SRGAN for image enhancement and embedding CBAM attention mechanism improves the accuracy of license plate image.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82574818","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00092
Haibo Ge, Haodong Feng, lJiajun Geng, Wenhao He, Yu An, Xing Song
With the rapid development of 5th Generation Mobile Communication Technology (5G), mobile edge computing (MEC) has played an important role in improving user experience and reducing energy consumption and latency. Mobile edge computing will decentralize some of the computing and storage problems of the central cloud to the network edge, so that the data generated by terminal devices can be processed quickly, but how to ensure that users can obtain good performance when moving between different locations is a key issue. In order to solve this problem and reduce end-to-end delay and energy consumption, a service migration algorithm based on node residual energy hybrid migration strategy for mobile edge computing (HOS-RE) is proposed. The experimental results show that compared with other methods, this method can reduce the time of service migration and ensure the continuity of services.
{"title":"Dynamic Service Migration Method Based on User Mobility","authors":"Haibo Ge, Haodong Feng, lJiajun Geng, Wenhao He, Yu An, Xing Song","doi":"10.1109/ICNLP58431.2023.00092","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00092","url":null,"abstract":"With the rapid development of 5th Generation Mobile Communication Technology (5G), mobile edge computing (MEC) has played an important role in improving user experience and reducing energy consumption and latency. Mobile edge computing will decentralize some of the computing and storage problems of the central cloud to the network edge, so that the data generated by terminal devices can be processed quickly, but how to ensure that users can obtain good performance when moving between different locations is a key issue. In order to solve this problem and reduce end-to-end delay and energy consumption, a service migration algorithm based on node residual energy hybrid migration strategy for mobile edge computing (HOS-RE) is proposed. The experimental results show that compared with other methods, this method can reduce the time of service migration and ensure the continuity of services.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90348181","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 : 2023-03-01DOI: 10.1109/ICNLP58431.2023.00023
Yu Lei, Guangyuan Zhao, Lingjie Zhang
Vegetation classification has a pivotal role in forest management and ecological research. It is a specific application problem in hyperspectral image classification. However, the existing classification models do not make sufficient use of the spatial features of vegetation, and cannot extract deep feature information. To address these issues, we propose a deep composite kernel extreme learning machine based on spatial feature extraction (DCKELM-SPATIAL) to classify vegetation. Especially, we use the Gabor filter and super-pixel density peak clustering method to obtain a new set of spatial composite kernels. Experiments are carried out on two sets of real hyperspectral vegetation datasets. The results show that this method is superior to some classical and advanced methods in classification accuracy, and satisfactory results are obtained.
{"title":"Deep Composite Kernels ELM Based on Spatial Feature Extraction for Hyperspectral Vegetation Image Classification","authors":"Yu Lei, Guangyuan Zhao, Lingjie Zhang","doi":"10.1109/ICNLP58431.2023.00023","DOIUrl":"https://doi.org/10.1109/ICNLP58431.2023.00023","url":null,"abstract":"Vegetation classification has a pivotal role in forest management and ecological research. It is a specific application problem in hyperspectral image classification. However, the existing classification models do not make sufficient use of the spatial features of vegetation, and cannot extract deep feature information. To address these issues, we propose a deep composite kernel extreme learning machine based on spatial feature extraction (DCKELM-SPATIAL) to classify vegetation. Especially, we use the Gabor filter and super-pixel density peak clustering method to obtain a new set of spatial composite kernels. Experiments are carried out on two sets of real hyperspectral vegetation datasets. The results show that this method is superior to some classical and advanced methods in classification accuracy, and satisfactory results are obtained.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"72693937","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 : 2023-03-01DOI: 10.1109/icnlp58431.2023.00009
Qiang Li, Haibo Ge, Chaofeng Huang, Ting Zhou
In view of the problem of missed detection and false detection in complex environment, especially at night environment and false target environment, the detection ability of target detection is poor. With the development of deep learning, an improved SSD-based target detection algorithm is proposed, and the attention mechanism and function fusion module are added on the basis of SSD, which is integrated into the original network. Secondly, FPN module is a kind of shallow network, which is used to integrate deep network and shallow network to improve the representation ability of semantic information. Experiments were carried out on VOC2007 data set, pseudo target detection data set and night target detection data set. The results show that the detection accuracy of this method is up to 92.1%, which is verified by the camouflage data set and the night target detection data set. Compared with SSD andMobile-V2-SSD, the detection accuracy of this method is improved by 16.3% and 4.8%, respectively, and it has better robustness and real-time detection ability in complex environments.
{"title":"Research based on improved SSD target detection algorithm","authors":"Qiang Li, Haibo Ge, Chaofeng Huang, Ting Zhou","doi":"10.1109/icnlp58431.2023.00009","DOIUrl":"https://doi.org/10.1109/icnlp58431.2023.00009","url":null,"abstract":"In view of the problem of missed detection and false detection in complex environment, especially at night environment and false target environment, the detection ability of target detection is poor. With the development of deep learning, an improved SSD-based target detection algorithm is proposed, and the attention mechanism and function fusion module are added on the basis of SSD, which is integrated into the original network. Secondly, FPN module is a kind of shallow network, which is used to integrate deep network and shallow network to improve the representation ability of semantic information. Experiments were carried out on VOC2007 data set, pseudo target detection data set and night target detection data set. The results show that the detection accuracy of this method is up to 92.1%, which is verified by the camouflage data set and the night target detection data set. Compared with SSD andMobile-V2-SSD, the detection accuracy of this method is improved by 16.3% and 4.8%, respectively, and it has better robustness and real-time detection ability in complex environments.","PeriodicalId":53637,"journal":{"name":"Icon","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73690632","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：