Machine learning techniques have been utilized on gene expression profiling for cancer diagnosis. However, the gene expression data suffer from the curse of high dimensionality. Different kinds of feature selection methods were proposed to decrease the features of specific cancer diagnosis. As the difficult of obtaining the samples of a particular tumor, the lack of training samples leads to the overfitting problem. To handle the two problems, this paper proposes an unsupervised feature learning method. This method is able to enhance the performance of unsupervised feature learning by leveraging the unlabeled samples from other sources. Since the method utilizes the knowledge among the expression data from different sources, it can boost cancer classification performance. The experimental results on the gene expression data proves that our method improves the generalization cancer diagnosis when the unlabeled data are used for unsupervised feature learning.
{"title":"An Unsupervised Feature Learning Method for Enhancing the Generalization of Cancer Diagnosis","authors":"Zhen Liu, Ruoyu Wang, Wen-bo Zhang, Deyu Tang","doi":"10.1145/3457682.3457720","DOIUrl":"https://doi.org/10.1145/3457682.3457720","url":null,"abstract":"Machine learning techniques have been utilized on gene expression profiling for cancer diagnosis. However, the gene expression data suffer from the curse of high dimensionality. Different kinds of feature selection methods were proposed to decrease the features of specific cancer diagnosis. As the difficult of obtaining the samples of a particular tumor, the lack of training samples leads to the overfitting problem. To handle the two problems, this paper proposes an unsupervised feature learning method. This method is able to enhance the performance of unsupervised feature learning by leveraging the unlabeled samples from other sources. Since the method utilizes the knowledge among the expression data from different sources, it can boost cancer classification performance. The experimental results on the gene expression data proves that our method improves the generalization cancer diagnosis when the unlabeled data are used for unsupervised feature learning.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"91 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123163580","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}
Boyu He, Han Wu, Congduan Li, Linqi Song, Weigang Chen
Semantic role labeling (SRL) is widely used to extract predicate-argument pairs from sentences. Traditional SRL methods can perform well on the single sentence but fail to work in dialogue scenario where ellipsis and anaphora frequently occurs. Some research work has been proposed to solve this problem, i.e. Conversational Semantic Role Labeling (CSRL), but there are still huge room for improvements. The error case study of BERT-based CSRL model has shown that the majority of the errors are observed in boundary matching, especially in entity mention detection. We think the premier cause of this kind of error is the deficiency of external knowledge such that the ill-informed model cannot correctly capture and correlate the entities. To this end, we propose to incorporate external knowledge into BERT using visible masking strategy. We evaluate our proposed model on DuConv dataset. Experimental results show that our model with knowledge enhancement outperforms the benchmarks. Further analysis also demonstrates that dialogue SRL can benefit from external knowledge.
{"title":"K-CSRL: Knowledge Enhanced Conversational Semantic Role Labeling","authors":"Boyu He, Han Wu, Congduan Li, Linqi Song, Weigang Chen","doi":"10.1145/3457682.3457763","DOIUrl":"https://doi.org/10.1145/3457682.3457763","url":null,"abstract":"Semantic role labeling (SRL) is widely used to extract predicate-argument pairs from sentences. Traditional SRL methods can perform well on the single sentence but fail to work in dialogue scenario where ellipsis and anaphora frequently occurs. Some research work has been proposed to solve this problem, i.e. Conversational Semantic Role Labeling (CSRL), but there are still huge room for improvements. The error case study of BERT-based CSRL model has shown that the majority of the errors are observed in boundary matching, especially in entity mention detection. We think the premier cause of this kind of error is the deficiency of external knowledge such that the ill-informed model cannot correctly capture and correlate the entities. To this end, we propose to incorporate external knowledge into BERT using visible masking strategy. We evaluate our proposed model on DuConv dataset. Experimental results show that our model with knowledge enhancement outperforms the benchmarks. Further analysis also demonstrates that dialogue SRL can benefit from external knowledge.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"78 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115743407","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}
Non-Intrusive Load Monitoring (NILM) has received widespread attention as an energy-saving technology. The method based on Hidden Markov Model (HMM) is very popular in this domain because of its relatively small demand for computing resources. However, the traditional HMM-based methods need additional information such as the working states of appliance to train the model. In this paper, we proposed a non-parameter model (IC-FHMM) to alleviate the problem that require prior knowledge. Experiments are conducted on three open-access datasets, and the results indicate that the proposed model is superior to the four state-of-the-art models on the metrics of Accuracy and F-measure.
{"title":"An Extended Factorial Hidden Markov Model for Non-Intrusive Load Monitoring Based on Density Peak Clustering","authors":"Zhao Wu, Chao Wang, Ruiyou Li, Huaiqing Zhang","doi":"10.1145/3457682.3457712","DOIUrl":"https://doi.org/10.1145/3457682.3457712","url":null,"abstract":"Non-Intrusive Load Monitoring (NILM) has received widespread attention as an energy-saving technology. The method based on Hidden Markov Model (HMM) is very popular in this domain because of its relatively small demand for computing resources. However, the traditional HMM-based methods need additional information such as the working states of appliance to train the model. In this paper, we proposed a non-parameter model (IC-FHMM) to alleviate the problem that require prior knowledge. Experiments are conducted on three open-access datasets, and the results indicate that the proposed model is superior to the four state-of-the-art models on the metrics of Accuracy and F-measure.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"2007 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125572812","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}
For classification tasks, probabilistic graphical models are usually categorized into two disjoint classes: generative or discriminative. It depends on the posterior probability p(x|y) of the label x given the observation y computation. On the one hand, generative models, like the Naive Bayes or the Hidden Markov Model (HMM), need the computation of the joint probability p(x, y), before using the Bayes rule to compute p(x|y). On the other hand, discriminative models compute p(x|y) directly, regardless of the observations’ law. They are intensively used nowadays, with models as Logistic Regression or Conditional Random Fields (CRF). However, the recent Entropic Forward-Backward algorithm shows that the HMM, considered as a generative model, can also match the discriminative one’s definition. This example leads to question if it is the case for other generative models. In this paper, we show that the Naive Bayes can also match the discriminative model definition, so it can be used in either a generative or a discriminative way. Moreover, this observation also discusses the notion of Generative-Discriminative pairs, linking, for example, Naive Bayes and Logistic Regression, or HMM and CRF. Related to this point, we show that the Logistic Regression can be viewed as a particular case of the Naive Bayes used in a discriminative way.
{"title":"Using the Naive Bayes as a discriminative model","authors":"E. Azeraf, E. Monfrini, W. Pieczynski","doi":"10.1145/3457682.3457697","DOIUrl":"https://doi.org/10.1145/3457682.3457697","url":null,"abstract":"For classification tasks, probabilistic graphical models are usually categorized into two disjoint classes: generative or discriminative. It depends on the posterior probability p(x|y) of the label x given the observation y computation. On the one hand, generative models, like the Naive Bayes or the Hidden Markov Model (HMM), need the computation of the joint probability p(x, y), before using the Bayes rule to compute p(x|y). On the other hand, discriminative models compute p(x|y) directly, regardless of the observations’ law. They are intensively used nowadays, with models as Logistic Regression or Conditional Random Fields (CRF). However, the recent Entropic Forward-Backward algorithm shows that the HMM, considered as a generative model, can also match the discriminative one’s definition. This example leads to question if it is the case for other generative models. In this paper, we show that the Naive Bayes can also match the discriminative model definition, so it can be used in either a generative or a discriminative way. Moreover, this observation also discusses the notion of Generative-Discriminative pairs, linking, for example, Naive Bayes and Logistic Regression, or HMM and CRF. Related to this point, we show that the Logistic Regression can be viewed as a particular case of the Naive Bayes used in a discriminative way.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"60 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129468671","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}
Hyperspectral images (HSI) have abundant bands and can capture more useful information, having been widely used in military and civil applications. Traditional HSI classification algorithms failed to take full consideration of the relationship between spatial-wise and spectral-wise information. In this paper, we propose the Spectral-wise Attention-based Residual Network (SARN), in which double branches structure is applied for HSI classification. There are two channels in the model. In the first channel, a novel spectral attention block is used to generate the attention map for the spectral-wise information. Then in the second channel, a spatial-wise residual unit is utilized to draw spatial features. Afterward, the spectral attention map and the spatial features are fused for classification. Experiment results on the Pavia University dataset and Indian_pines dataset demonstrate that the proposed method has better performance than the state-of-art method.
{"title":"Spectral-wise Attention-based Residual Network for Hyperspectral Image Classification","authors":"Yaxin Chen, Zhiqiang Guo, Jie Yang","doi":"10.1145/3457682.3457735","DOIUrl":"https://doi.org/10.1145/3457682.3457735","url":null,"abstract":"Hyperspectral images (HSI) have abundant bands and can capture more useful information, having been widely used in military and civil applications. Traditional HSI classification algorithms failed to take full consideration of the relationship between spatial-wise and spectral-wise information. In this paper, we propose the Spectral-wise Attention-based Residual Network (SARN), in which double branches structure is applied for HSI classification. There are two channels in the model. In the first channel, a novel spectral attention block is used to generate the attention map for the spectral-wise information. Then in the second channel, a spatial-wise residual unit is utilized to draw spatial features. Afterward, the spectral attention map and the spatial features are fused for classification. Experiment results on the Pavia University dataset and Indian_pines dataset demonstrate that the proposed method has better performance than the state-of-art method.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127461367","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}
Avelumab, approved by the US Food and Drug Administration (FDA) for the treatment of Merkel cell carcinoma in adults and paediatric patients in 2017, is an investigational fully human anti–PD-L1 IgG1 antibody that inhibits PD-1/PD-L1 interactions. Although the crystal structure of the avelumab/PD-L1 complex was reported in 2017, which provided us the interface information at atom level, the dynamics information of the complex is missed, and some key residues could not be detected in that static crystal structure. Here, molecular dynamics simulations were performed for the avelumab/PD-L1 complex to map the epitope to paratope residues. The results showed that the epitope residues locating on the C strand (PD-L1TYR56 and PD-L1GLU58), CC’ loop (PD-L1GLU60, PD-L1ASP61 and PD-L1LYS62), C’ strand (PD-L1ASN63), and C'D loop (PD-L1HIS69) of PD-L1 mainly form the interface with avelumab. The paratope residues on avelumab include TYR52H, SER54H, GLY102H, THR105H, TYR34L, ASP52L and ARG99L. The C’ strand of PD-L1 is also a binding region for PD-1. Thus, antibody avelumab block PD-1/PD-L1 interaction through direct competitive binding of the C’ strand of PD-L1.
{"title":"Identifying the Key Residues Regulating the Binding between Antibody Avelumab and PD-L1 VIA Molecular Dynamics Simulation","authors":"Wenping Liu, Ting Chen, Shengsheng Lai, Gangping Zhang, Guangjian Liu, Haoyu Jin","doi":"10.1145/3457682.3457767","DOIUrl":"https://doi.org/10.1145/3457682.3457767","url":null,"abstract":"Avelumab, approved by the US Food and Drug Administration (FDA) for the treatment of Merkel cell carcinoma in adults and paediatric patients in 2017, is an investigational fully human anti–PD-L1 IgG1 antibody that inhibits PD-1/PD-L1 interactions. Although the crystal structure of the avelumab/PD-L1 complex was reported in 2017, which provided us the interface information at atom level, the dynamics information of the complex is missed, and some key residues could not be detected in that static crystal structure. Here, molecular dynamics simulations were performed for the avelumab/PD-L1 complex to map the epitope to paratope residues. The results showed that the epitope residues locating on the C strand (PD-L1TYR56 and PD-L1GLU58), CC’ loop (PD-L1GLU60, PD-L1ASP61 and PD-L1LYS62), C’ strand (PD-L1ASN63), and C'D loop (PD-L1HIS69) of PD-L1 mainly form the interface with avelumab. The paratope residues on avelumab include TYR52H, SER54H, GLY102H, THR105H, TYR34L, ASP52L and ARG99L. The C’ strand of PD-L1 is also a binding region for PD-1. Thus, antibody avelumab block PD-1/PD-L1 interaction through direct competitive binding of the C’ strand of PD-L1.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126837932","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}
Deep learning algorithms perform well in the field of object detection. Object detection networks represented by YOLO, SSD and faster-RCNN have achieved excellent performance on public datasets such as VOC and COCO. However, deep learning models are difficult to deploy on the edge computing platform with less computing resources due to its huge amount of parameters and computation. In this paper, we propose an efficient dynamic sparsity method to help the network quickly mine important parameters, and then prune the unimportant weight channels, which makes the network model more compact and consumes less computation. In the case of high sparsity, our method is more robust than L1 regularization and other regularization forms, and can achieve better sparsity and pruning effects. Through this method, we can prune the YOLOv3 network and the enhanced YOLOv3-SPP3 network by up to 90%. This allows the network to achieve 5× reduction in FLOPs and maintain an accuracy loss of less than 1% on the BDD100k dataset.
{"title":"YOLO-Tight: an Efficient Dynamic Compression Method for YOLO Object Detection Networks","authors":"Wei Yan, Ting Liu, Yuzhuo Fu","doi":"10.1145/3457682.3457740","DOIUrl":"https://doi.org/10.1145/3457682.3457740","url":null,"abstract":"Deep learning algorithms perform well in the field of object detection. Object detection networks represented by YOLO, SSD and faster-RCNN have achieved excellent performance on public datasets such as VOC and COCO. However, deep learning models are difficult to deploy on the edge computing platform with less computing resources due to its huge amount of parameters and computation. In this paper, we propose an efficient dynamic sparsity method to help the network quickly mine important parameters, and then prune the unimportant weight channels, which makes the network model more compact and consumes less computation. In the case of high sparsity, our method is more robust than L1 regularization and other regularization forms, and can achieve better sparsity and pruning effects. Through this method, we can prune the YOLOv3 network and the enhanced YOLOv3-SPP3 network by up to 90%. This allows the network to achieve 5× reduction in FLOPs and maintain an accuracy loss of less than 1% on the BDD100k dataset.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128175736","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}
Dagmawi Alemu Moges, Andre Niyongabo Rubungo, Hong Qu
Machine Reading Comprehension is defined as the ability of machines to read and understand unstructured text and answer questions about it. It is considered as a challenging task with wide range of enterprise applications. Wide range of natural language understanding and reasoning tasks are found embedded within machine reading comprehension datasets. This requires effective models with robust relational reasoning capabilities to answer complex questions. Reasoning in natural language is a long-term machine-learning goal and is critically needed for building intelligent agents. However, most papers heavily depend on underlying language modeling and thus pay little to no attention on creating effective reasoning models. This paper proposes a modified transformer architecture that effectively combines soft and hard attention to create multi-perspective reasoning model capable of tackling wide range of reasoning tasks. An attention mechanism that highlights the relational significance of input signals is considered as well. The result from this study shows performance gain as compared to its counterpart the transformer network on bAbI dataset, a natural language reasoning tasks.
{"title":"Multi-Perspective Reasoning Transformers","authors":"Dagmawi Alemu Moges, Andre Niyongabo Rubungo, Hong Qu","doi":"10.1145/3457682.3457759","DOIUrl":"https://doi.org/10.1145/3457682.3457759","url":null,"abstract":"Machine Reading Comprehension is defined as the ability of machines to read and understand unstructured text and answer questions about it. It is considered as a challenging task with wide range of enterprise applications. Wide range of natural language understanding and reasoning tasks are found embedded within machine reading comprehension datasets. This requires effective models with robust relational reasoning capabilities to answer complex questions. Reasoning in natural language is a long-term machine-learning goal and is critically needed for building intelligent agents. However, most papers heavily depend on underlying language modeling and thus pay little to no attention on creating effective reasoning models. This paper proposes a modified transformer architecture that effectively combines soft and hard attention to create multi-perspective reasoning model capable of tackling wide range of reasoning tasks. An attention mechanism that highlights the relational significance of input signals is considered as well. The result from this study shows performance gain as compared to its counterpart the transformer network on bAbI dataset, a natural language reasoning tasks.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129940334","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}
Partial differential equations (PDEs) are the most ubiquitous tool for modeling problems in nature. In recent years, machine learning techniques are adopted to solve PDEs. However, the prediction errors of existing machine learning methods vary widely on different subdomains of PDEs. How to achieve precision-consistency is a crucial and complex issue for machine learning methods for solving PDEs. To tackle this issue, we propose DSP, an adaptive framework for solving PDEs. DSP is composed of domain decomposition, searching for singular subdomains, and prediction. Furthermore, a novel generative model, physics-informed generative adversarial network (PIGAN), is designed to solve PDEs. In addition, we introduce points with high-precision labels into the training process of the model to improve model accuracy. We test the effectiveness of our approach on three real physical equations: Poisson equation, Helmhotz equation and Eikonal equation. Through experiments, we prove that the combination of DSP and PIGAN outperforms various state-of-the-art baselines.
{"title":"DSP-PIGAN: A Precision-Consistency Machine Learning Algorithm for Solving Partial Differential Equations","authors":"Yunzhuo Wang, Hao Sun, Guangzhong Sun","doi":"10.1145/3457682.3457686","DOIUrl":"https://doi.org/10.1145/3457682.3457686","url":null,"abstract":"Partial differential equations (PDEs) are the most ubiquitous tool for modeling problems in nature. In recent years, machine learning techniques are adopted to solve PDEs. However, the prediction errors of existing machine learning methods vary widely on different subdomains of PDEs. How to achieve precision-consistency is a crucial and complex issue for machine learning methods for solving PDEs. To tackle this issue, we propose DSP, an adaptive framework for solving PDEs. DSP is composed of domain decomposition, searching for singular subdomains, and prediction. Furthermore, a novel generative model, physics-informed generative adversarial network (PIGAN), is designed to solve PDEs. In addition, we introduce points with high-precision labels into the training process of the model to improve model accuracy. We test the effectiveness of our approach on three real physical equations: Poisson equation, Helmhotz equation and Eikonal equation. Through experiments, we prove that the combination of DSP and PIGAN outperforms various state-of-the-art baselines.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133447795","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}
Images captured by Unmanned Surface Vessel (USV) have a wide range of applications in various fields, such as maritime object detection, remote sensing, and autonomous transportation. However, cameras often suffer from a low light environment, resulting in low contrast, high noise, and poor quality image, causing identification difficulties and machine decision errors. In recent years, convolutional neural networks have developed rapidly, which have strong generalization ability and can extract different levels of information, especially high-level information. Therefore, to preprocess low light images before advanced computer vision tasks of USV, we proposed a deep learning-based end-to-end convolutional network for low light enhancement in USV imaging system. The advantage of our model is using U-Net as the basic architecture to gain multi-scale feature maps with improvements, including attention mechanism and dense connection. Besides, we pay attention to edge information given images' edge loss. With the unique network structure, our model can effectively increase the brightness and contrast of dark aquatic images. Experiments have been carried out on testing images to analyze our proposed method with several latest imaging methods. The experimental results show its outstanding performance in both subjective and objective evaluation.
{"title":"Low Light Image Enhancement in USV Imaging System Via U-Net and Attention Mechanism","authors":"Sheng Zhang, Tianxiao Cai, Yihang Chen","doi":"10.1145/3457682.3457729","DOIUrl":"https://doi.org/10.1145/3457682.3457729","url":null,"abstract":"Images captured by Unmanned Surface Vessel (USV) have a wide range of applications in various fields, such as maritime object detection, remote sensing, and autonomous transportation. However, cameras often suffer from a low light environment, resulting in low contrast, high noise, and poor quality image, causing identification difficulties and machine decision errors. In recent years, convolutional neural networks have developed rapidly, which have strong generalization ability and can extract different levels of information, especially high-level information. Therefore, to preprocess low light images before advanced computer vision tasks of USV, we proposed a deep learning-based end-to-end convolutional network for low light enhancement in USV imaging system. The advantage of our model is using U-Net as the basic architecture to gain multi-scale feature maps with improvements, including attention mechanism and dense connection. Besides, we pay attention to edge information given images' edge loss. With the unique network structure, our model can effectively increase the brightness and contrast of dark aquatic images. Experiments have been carried out on testing images to analyze our proposed method with several latest imaging methods. The experimental results show its outstanding performance in both subjective and objective evaluation.","PeriodicalId":142045,"journal":{"name":"2021 13th International Conference on Machine Learning and Computing","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125186563","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：