Pub Date : 2019-03-18DOI: 10.1109/ICAIIC.2019.8669016
Ghudae Sim, Hyungbin Yun, Junhee Seok
While it is popular, estimating empirical distribution from observed data using MSE (Mean Squared Error) is often inefficient because it focuses on expectation. To address this problem, here we invest a new type of error term, named MRE (Mean Root Error). Different from MSE, MRE can predict the local mode point rather than the expectation. From numerical studies, we show that MRE models shows more robust and accurate prediction performance, which will be useful for complicated data such as finance data.
{"title":"The Properties of mode prediction using mean root error for regularization","authors":"Ghudae Sim, Hyungbin Yun, Junhee Seok","doi":"10.1109/ICAIIC.2019.8669016","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8669016","url":null,"abstract":"While it is popular, estimating empirical distribution from observed data using MSE (Mean Squared Error) is often inefficient because it focuses on expectation. To address this problem, here we invest a new type of error term, named MRE (Mean Root Error). Different from MSE, MRE can predict the local mode point rather than the expectation. From numerical studies, we show that MRE models shows more robust and accurate prediction performance, which will be useful for complicated data such as finance data.","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133383301","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 : 2019-03-18DOI: 10.1109/ICAIIC.2019.8669028
Jing Ma, T. Nagatsuma, Song-Ju Kim, M. Hasegawa
Multi-channel technique benefits IoT network by support parallel transmission and reduce interference. However, the extra overhead posed by the multi-channel usage coordination dramatically challenges the resource constrained IoT devices. In this paper, a machine-learning-based channel assignment algorithm utilizing Tug-Of-War (TOW) dynamics is proposed to cognitively select channels for communication in massive IoT. Furthermore, the proposed TOW-dynamics-based channel assignment algorithm has simple learning procedure which only needs to receive Acknowledge frame for learning procedure, meanwhile, only needs minimal memory and computation capability, i.e., addition and subtraction procedure. Thus, the proposed TOW-dynamics-based algorithm is possible to run on resource constrained IoT devices. We prototype the proposed algorithm on extremely resource constrained Single-board Computer, which is called cognitive IoT device hereafter. Moreover, the evaluation experiments that densely deployed cognitive IoT devices in the frequently changed radio environment are conducted. The evaluation results show that cognitive IoT device quickly make decision to selects channel when the real environment frequently changed, meanwhile keep fairness among IoT devices.
{"title":"A Machine-Learning-Based Channel Assignment Algorithm for IoT","authors":"Jing Ma, T. Nagatsuma, Song-Ju Kim, M. Hasegawa","doi":"10.1109/ICAIIC.2019.8669028","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8669028","url":null,"abstract":"Multi-channel technique benefits IoT network by support parallel transmission and reduce interference. However, the extra overhead posed by the multi-channel usage coordination dramatically challenges the resource constrained IoT devices. In this paper, a machine-learning-based channel assignment algorithm utilizing Tug-Of-War (TOW) dynamics is proposed to cognitively select channels for communication in massive IoT. Furthermore, the proposed TOW-dynamics-based channel assignment algorithm has simple learning procedure which only needs to receive Acknowledge frame for learning procedure, meanwhile, only needs minimal memory and computation capability, i.e., addition and subtraction procedure. Thus, the proposed TOW-dynamics-based algorithm is possible to run on resource constrained IoT devices. We prototype the proposed algorithm on extremely resource constrained Single-board Computer, which is called cognitive IoT device hereafter. Moreover, the evaluation experiments that densely deployed cognitive IoT devices in the frequently changed radio environment are conducted. The evaluation results show that cognitive IoT device quickly make decision to selects channel when the real environment frequently changed, meanwhile keep fairness among IoT devices.","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"29 5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129997581","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 : 2019-03-18DOI: 10.1109/ICAIIC.2019.8668996
Hyungbin Yun, Ghudae Sim, Junhee Seok
Non-quantitative data have a significant impact on the financial market as well as quantitative data. In this paper, we propose CNN model of stock price prediction using Korean natural language processing. In the case of Korean natural language processing research was not actively performed compared to English language. We converted Korean sentences into nouns and vectorized them using skip-grams to extract the characteristics of the words. Then, the vectorized word sentence was used as input data of the CNN model to predict the stock price after 5 days of trading day. Most models have more than 50% prediction accuracy for stock price up and down. The highest accuracy of the model was about 53%. Since the result is not considerable but meaningful, it shows the possibility of developing the stock price prediction model through Korean natural language processing in the future.
{"title":"Stock Prices Prediction using the Title of Newspaper Articles with Korean Natural Language Processing","authors":"Hyungbin Yun, Ghudae Sim, Junhee Seok","doi":"10.1109/ICAIIC.2019.8668996","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8668996","url":null,"abstract":"Non-quantitative data have a significant impact on the financial market as well as quantitative data. In this paper, we propose CNN model of stock price prediction using Korean natural language processing. In the case of Korean natural language processing research was not actively performed compared to English language. We converted Korean sentences into nouns and vectorized them using skip-grams to extract the characteristics of the words. Then, the vectorized word sentence was used as input data of the CNN model to predict the stock price after 5 days of trading day. Most models have more than 50% prediction accuracy for stock price up and down. The highest accuracy of the model was about 53%. Since the result is not considerable but meaningful, it shows the possibility of developing the stock price prediction model through Korean natural language processing in the future.","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127223428","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 : 2019-03-18DOI: 10.1109/ICAIIC.2019.8669030
Hoesang Choi, H. Moon
To reduce power consumption in random access, channel-adaptive random access was suggested. With channel-adaptive random access, a remote station transmits a random access packet only when channel gain is greater than a predetermined threshold. Even thought a random access event is triggered, if channel gain is less than the threshold, the remote station waits to transmit the packet until the channel gain becomes greater than the threshold. Therefore, there is an additional delay compared to conventional random access. Previous researches showed that channel-adaptive random access has a trade-off between power consumption and delay. However, retransmission was not considered in the previous researches. Therefore, in this paper, random access delay is compared between conventional and channel-adaptive random access considering retransmissions.
{"title":"Simulation on Delay of Several Random Access Schemes","authors":"Hoesang Choi, H. Moon","doi":"10.1109/ICAIIC.2019.8669030","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8669030","url":null,"abstract":"To reduce power consumption in random access, channel-adaptive random access was suggested. With channel-adaptive random access, a remote station transmits a random access packet only when channel gain is greater than a predetermined threshold. Even thought a random access event is triggered, if channel gain is less than the threshold, the remote station waits to transmit the packet until the channel gain becomes greater than the threshold. Therefore, there is an additional delay compared to conventional random access. Previous researches showed that channel-adaptive random access has a trade-off between power consumption and delay. However, retransmission was not considered in the previous researches. Therefore, in this paper, random access delay is compared between conventional and channel-adaptive random access considering retransmissions.","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129034513","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 : 2019-03-18DOI: 10.1109/ICAIIC.2019.8669055
Pilsub Lee, Myungwon Choi, Daegyeom Kim, Suji Lee, Hyun-Ghang Jeong, C. E. Han
A brain network is the essence of the intelligence where it consists of nodes that are anatomically defined brain regions, and edges that connect a pair of brain regions. The diffusion-weighted magnetic resonance (MR) images and the advances in computer-aided tractography algorithms let us know strong association between human brain networks and cognitive functions. Brain regions dedicated to a certain specific cognitive function were spatially clustered and efficiently connected each other; it is called local functional segregation. However, it is not well known that such a local segregation is associated with a certain sub-network which may act as a building block of the brain network. In this work, using a graph auto-encoder, we extracted building blocks of brain networks and investigate whether they are affected by a neurological disease, Alzheimer’s disease. We found that the brain network of each person is linear summation of the learned building blocks. Also, the activation levels of these building blocks vary in the normal controls and patients with Alzheimer’s disease, showing that network deterioration in the disease group.
{"title":"Deep learning based decomposition of brain networks","authors":"Pilsub Lee, Myungwon Choi, Daegyeom Kim, Suji Lee, Hyun-Ghang Jeong, C. E. Han","doi":"10.1109/ICAIIC.2019.8669055","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8669055","url":null,"abstract":"A brain network is the essence of the intelligence where it consists of nodes that are anatomically defined brain regions, and edges that connect a pair of brain regions. The diffusion-weighted magnetic resonance (MR) images and the advances in computer-aided tractography algorithms let us know strong association between human brain networks and cognitive functions. Brain regions dedicated to a certain specific cognitive function were spatially clustered and efficiently connected each other; it is called local functional segregation. However, it is not well known that such a local segregation is associated with a certain sub-network which may act as a building block of the brain network. In this work, using a graph auto-encoder, we extracted building blocks of brain networks and investigate whether they are affected by a neurological disease, Alzheimer’s disease. We found that the brain network of each person is linear summation of the learned building blocks. Also, the activation levels of these building blocks vary in the normal controls and patients with Alzheimer’s disease, showing that network deterioration in the disease group.","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128286101","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 : 2019-02-13DOI: 10.1109/ICAIIC.2019.8669034
Hyun Kwon, H. Yoon, D. Choi
Deep neural networks (DNNs) provide superior per-formance on machine learning tasks such as image recognition, speech recognition, pattern recognition, and intrusion detection. However, an adversarial example created by adding a little noise to the original data can lead to misclassification by the DNN, and the human eye cannot detect the difference from the original data. For example, if an attacker generates a modified left-turn road sign to be incorrectly categorized by a DNN, an autonomous vehicle with the DNN will incorrect classify the modified left-turn road sign as a right-turn sign, whereas a human will correctly classify the modified sign as a left-turn sign. Such an adversarial example is a serious threat to a DNN. Recently, a multi-target adversarial example was introduced that causes misclassification by several models within each target class using a single modified image. However, it has the vulnerability that as the number of target models increases, the overall attack success rate is reduced. Therefore, if there are several models that the attacker wishes to target, the attacker needs to control the attack success rate for each model by considering the attack priority for each model. In this paper, we propose a priority adversarial example that considers the attack priority for each model in cases targeting several models. The proposed method controls the attack success rate for each model by adjusting the weight of the attack function in the generation process, while maintaining minimum distortion. We used Tensorflow, a widely used machine learning library, and MNIST as the dataset. Experimental results show that the proposed method can control the attack success rate for each model by considering the attack priority of each model while maintaining minimum distortion (on average 3.95 and 2.45 in targeted and untargeted attacks, respectively).
{"title":"Priority Adversarial Example in Evasion Attack on Multiple Deep Neural Networks","authors":"Hyun Kwon, H. Yoon, D. Choi","doi":"10.1109/ICAIIC.2019.8669034","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8669034","url":null,"abstract":"Deep neural networks (DNNs) provide superior per-formance on machine learning tasks such as image recognition, speech recognition, pattern recognition, and intrusion detection. However, an adversarial example created by adding a little noise to the original data can lead to misclassification by the DNN, and the human eye cannot detect the difference from the original data. For example, if an attacker generates a modified left-turn road sign to be incorrectly categorized by a DNN, an autonomous vehicle with the DNN will incorrect classify the modified left-turn road sign as a right-turn sign, whereas a human will correctly classify the modified sign as a left-turn sign. Such an adversarial example is a serious threat to a DNN. Recently, a multi-target adversarial example was introduced that causes misclassification by several models within each target class using a single modified image. However, it has the vulnerability that as the number of target models increases, the overall attack success rate is reduced. Therefore, if there are several models that the attacker wishes to target, the attacker needs to control the attack success rate for each model by considering the attack priority for each model. In this paper, we propose a priority adversarial example that considers the attack priority for each model in cases targeting several models. The proposed method controls the attack success rate for each model by adjusting the weight of the attack function in the generation process, while maintaining minimum distortion. We used Tensorflow, a widely used machine learning library, and MNIST as the dataset. Experimental results show that the proposed method can control the attack success rate for each model by considering the attack priority of each model while maintaining minimum distortion (on average 3.95 and 2.45 in targeted and untargeted attacks, respectively).","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"329 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-02-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123395437","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 : 2019-02-01DOI: 10.1109/ICAIIC.2019.8669022
M. Kikuchi
In this paper, we introduce a method to estimate the control parameters of the neuromuscular skeletal system by using convolution neural networks (CNNs) and parametric identification together. We selected a human standing attitude control system as an object and conducted measurement experiments with the output of barycenter fluctuation with visual instructions as input. Then, we created an image emphasizing the features of the result and carried out transfer learning with stochastic gradient descent method with CNNs using the AlexNet. Here, we classified the measurement time and the result for each subject into 16 classes. On the other hand, we created a mathematical model of the system and carried out a parametric identification of the control parameters of the neuromuscular skeletal system the standing attitude control system using the ARMAX algorithm. Next, using the feature extraction method, the CNN output, and the original parameter estimation method, the control parameters of the verification data estimated without using ARMAX from the CNN output and the control parameter information of the learning data. As the results, (1) the center of gravity fluctuation differs for each subject and each hour. (2) feature extraction works effectively. (3) It is possible to correlate data classified by CNNs and estimated values of control parameters. We showed these three results. Moreover, from the viewpoint of the system model, it suggested that the decision process of deep learning could be analyzed using the change of internal parameters in the system.
{"title":"Estimation of Control Parameters in Neuromuscular Skeletal Systems Combined with CNNs and Parametric Identification","authors":"M. Kikuchi","doi":"10.1109/ICAIIC.2019.8669022","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8669022","url":null,"abstract":"In this paper, we introduce a method to estimate the control parameters of the neuromuscular skeletal system by using convolution neural networks (CNNs) and parametric identification together. We selected a human standing attitude control system as an object and conducted measurement experiments with the output of barycenter fluctuation with visual instructions as input. Then, we created an image emphasizing the features of the result and carried out transfer learning with stochastic gradient descent method with CNNs using the AlexNet. Here, we classified the measurement time and the result for each subject into 16 classes. On the other hand, we created a mathematical model of the system and carried out a parametric identification of the control parameters of the neuromuscular skeletal system the standing attitude control system using the ARMAX algorithm. Next, using the feature extraction method, the CNN output, and the original parameter estimation method, the control parameters of the verification data estimated without using ARMAX from the CNN output and the control parameter information of the learning data. As the results, (1) the center of gravity fluctuation differs for each subject and each hour. (2) feature extraction works effectively. (3) It is possible to correlate data classified by CNNs and estimated values of control parameters. We showed these three results. Moreover, from the viewpoint of the system model, it suggested that the decision process of deep learning could be analyzed using the change of internal parameters in the system.","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115732248","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 : 2019-02-01DOI: 10.1109/ICAIIC.2019.8669076
Tae-Hong Min, Do-Yun Kim, Young-June Choi
Unsupervised image-to-image translation can map local textures between two domains, but typically fails when the domain requires big shape changes. It is difficult to learn how to make such big change using the basic convolution layer, and furthermore it takes much time to learn. For faster learning and high-quality image generation, we propose to use Cycle GAN that is combined with Resnet in a network that is connected with the residual block for upsampling to make big shape change and construct faster image-to-image translation.
{"title":"Improving Learning time in Unsupervised Image-to-Image Translation","authors":"Tae-Hong Min, Do-Yun Kim, Young-June Choi","doi":"10.1109/ICAIIC.2019.8669076","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8669076","url":null,"abstract":"Unsupervised image-to-image translation can map local textures between two domains, but typically fails when the domain requires big shape changes. It is difficult to learn how to make such big change using the basic convolution layer, and furthermore it takes much time to learn. For faster learning and high-quality image generation, we propose to use Cycle GAN that is combined with Resnet in a network that is connected with the residual block for upsampling to make big shape change and construct faster image-to-image translation.","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122771380","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 : 2019-02-01DOI: 10.1109/ICAIIC.2019.8669062
Hoan Nguyen, Minh Duc Thieu, Huy Nguyen, Tung Lam Pham, N. Le, Y. Jang
In this paper, we have introduced Optical Camera Communication, challenging and solution of RoI signal and S2-PSK modulation scheme. Nowadays, OCC has been established in many areas which also be used in vehicle communication. Although OCC has been the concern of a vehement research during the recent years, the technology is still in its juvenile and requires continuous efforts to overcome the current challenges, especially in outdoor applications, such as the vehicle communications.
{"title":"Solution for Sampling Time Deviation in Decoding RoI-Signaling Waveform Using S2-PSK","authors":"Hoan Nguyen, Minh Duc Thieu, Huy Nguyen, Tung Lam Pham, N. Le, Y. Jang","doi":"10.1109/ICAIIC.2019.8669062","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8669062","url":null,"abstract":"In this paper, we have introduced Optical Camera Communication, challenging and solution of RoI signal and S2-PSK modulation scheme. Nowadays, OCC has been established in many areas which also be used in vehicle communication. Although OCC has been the concern of a vehement research during the recent years, the technology is still in its juvenile and requires continuous efforts to overcome the current challenges, especially in outdoor applications, such as the vehicle communications.","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121935369","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 : 2019-02-01DOI: 10.1109/ICAIIC.2019.8669083
Gwangsun Kim, Junyeong Kim, Sunghwan Kim
Fire detection is very crucial to the security and important to preserve the properties of citizens. On fire detection, various features such as extracted information from video and others have been used. The combination of various features can improve the accuracy of fire detection. Usually video images are an important resource for this task, and prior knowledge about colors and variations of fires can be used. Recently, deep neural network has shown the best performance in many task in computer visions. Thus, the use of deep neural network in fire detection has risen, but there were little works to use the temporally summarized information from the prior knowledge. To construct the deep neural network architecture reflecting this information and validate its performances, we gathered video clips and proposed the deep neural network using the temporal information from video clips is proposed. Analysis of real data showed that the proposed method improve the accuracy significantly. To summarize the temporal information we use the standard deviation of G-filter values of images along the time. By using this information, the more compact architecture can be constructed.
{"title":"Fire Detection Using Video Images and Temporal Variations","authors":"Gwangsun Kim, Junyeong Kim, Sunghwan Kim","doi":"10.1109/ICAIIC.2019.8669083","DOIUrl":"https://doi.org/10.1109/ICAIIC.2019.8669083","url":null,"abstract":"Fire detection is very crucial to the security and important to preserve the properties of citizens. On fire detection, various features such as extracted information from video and others have been used. The combination of various features can improve the accuracy of fire detection. Usually video images are an important resource for this task, and prior knowledge about colors and variations of fires can be used. Recently, deep neural network has shown the best performance in many task in computer visions. Thus, the use of deep neural network in fire detection has risen, but there were little works to use the temporally summarized information from the prior knowledge. To construct the deep neural network architecture reflecting this information and validate its performances, we gathered video clips and proposed the deep neural network using the temporal information from video clips is proposed. Analysis of real data showed that the proposed method improve the accuracy significantly. To summarize the temporal information we use the standard deviation of G-filter values of images along the time. By using this information, the more compact architecture can be constructed.","PeriodicalId":273383,"journal":{"name":"2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"80 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117028081","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：