Pub Date : 2020-12-02DOI: 10.1109/ICMLC51923.2020.9469561
Cheng Xu, G. Ji, Bin Zhao
In recent years, urban traffic congestion has become increasingly serious. More and more scholars have begun to study about intelligent transportation system, and the real-time detection of vehicle flow is one of the most valuable research issues. In this paper, we propose an algorithm called VFDV (Vehicle Flow Detection algorithm based on Video) that can detect vehicle flow in real time. This algorithm uses road video surveillance as the source data and extracts valid images from it to detect vehicle flow. Different from the traditional methods that use vehicle recognition method to detect vehicle flow, algorithm VFDV uses a classification algorithm to detect vehicle flow. Compared with traditional algorithms, our algorithm achieves higher accuracy. In the verification phase, the video taken at the real intersection is used as the data source. Experiments on real dataset are designed to verify the effectiveness and superiority of the proposed algorithm.
近年来，城市交通拥堵问题日益严重。越来越多的学者开始对智能交通系统进行研究，其中车辆流量的实时检测是最有价值的研究问题之一。本文提出了一种基于视频的车辆流量检测算法VFDV (Vehicle Flow Detection algorithm based on Video)，可以实时检测车辆流量。该算法以道路视频监控为源数据，提取有效图像进行车辆流量检测。与传统的车辆识别检测车辆流量的方法不同，VFDV算法采用分类算法检测车辆流量。与传统算法相比，我们的算法达到了更高的精度。在验证阶段，使用在真实路口拍摄的视频作为数据源。在实际数据集上进行了实验，验证了该算法的有效性和优越性。
{"title":"Video-Based Vehicle Flow Detection Algorithm","authors":"Cheng Xu, G. Ji, Bin Zhao","doi":"10.1109/ICMLC51923.2020.9469561","DOIUrl":"https://doi.org/10.1109/ICMLC51923.2020.9469561","url":null,"abstract":"In recent years, urban traffic congestion has become increasingly serious. More and more scholars have begun to study about intelligent transportation system, and the real-time detection of vehicle flow is one of the most valuable research issues. In this paper, we propose an algorithm called VFDV (Vehicle Flow Detection algorithm based on Video) that can detect vehicle flow in real time. This algorithm uses road video surveillance as the source data and extracts valid images from it to detect vehicle flow. Different from the traditional methods that use vehicle recognition method to detect vehicle flow, algorithm VFDV uses a classification algorithm to detect vehicle flow. Compared with traditional algorithms, our algorithm achieves higher accuracy. In the verification phase, the video taken at the real intersection is used as the data source. Experiments on real dataset are designed to verify the effectiveness and superiority of the proposed algorithm.","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128087115","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 : 2020-12-02DOI: 10.1109/ICMLC51923.2020.9469580
Maisha Farzana, Md. Jahid Hossain Any, Md. Tanzim Reza, M. Parvez
Automated semantic segmentation of brain tumors from 3D MRI images plays a significant role in medical image processing, monitoring and diagnosis. Early detection of these brain tumors is highly requisite for the treatment, diagnosis and surgical pre-planning of the anomalies. The physicians normally follow the manual way of delineation for diagnosis of tumors which is time consuming and requires too much knowledge of anatomy. To resolve these limitations, convolutional neural network (CNN) based U-Net autoencoder model is proposed which performs automated segmentation of brain tumors from 3D MRI brain images by extracting the key features of the tumor. Additionally, Image normalization, image augmentation, image binarization etc. are applied for data pre-processing. Later on, the model is applied to the new 3D MRI brain images to test the accuracy of it. Applying the proposed method, the accuracy is obtained upto 96.06% considering the 18 subjects. Finally, this approach is a well-structured model for segmenting the tumor region from MRI brain images as compare to the other existing models which may assist the physicians for better diagnosis and therefore, opening the door for more precise therapy and better treatment to the patient.
{"title":"Semantic Segmentation of Brain Tumor from 3D Structural MRI Using U-Net Autoencoder","authors":"Maisha Farzana, Md. Jahid Hossain Any, Md. Tanzim Reza, M. Parvez","doi":"10.1109/ICMLC51923.2020.9469580","DOIUrl":"https://doi.org/10.1109/ICMLC51923.2020.9469580","url":null,"abstract":"Automated semantic segmentation of brain tumors from 3D MRI images plays a significant role in medical image processing, monitoring and diagnosis. Early detection of these brain tumors is highly requisite for the treatment, diagnosis and surgical pre-planning of the anomalies. The physicians normally follow the manual way of delineation for diagnosis of tumors which is time consuming and requires too much knowledge of anatomy. To resolve these limitations, convolutional neural network (CNN) based U-Net autoencoder model is proposed which performs automated segmentation of brain tumors from 3D MRI brain images by extracting the key features of the tumor. Additionally, Image normalization, image augmentation, image binarization etc. are applied for data pre-processing. Later on, the model is applied to the new 3D MRI brain images to test the accuracy of it. Applying the proposed method, the accuracy is obtained upto 96.06% considering the 18 subjects. Finally, this approach is a well-structured model for segmenting the tumor region from MRI brain images as compare to the other existing models which may assist the physicians for better diagnosis and therefore, opening the door for more precise therapy and better treatment to the patient.","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"40 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132706135","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 : 2020-12-02DOI: 10.1109/ICMLC51923.2020.9469587
Diptangshu Pandit, Li Zhang, Kamlesh Mistry, Richard M. Jiang
In this research, we propose a hybrid meta-classifier for novel class detection. It is able to efficiently detect the arrival of novel unseen classes as well as tackle real-time data stream classification. Specifically, the proposed hybrid meta-classifler includes three ensemble models, i.e. class-specific, cluster-specific and complementary boosting ensemble classifiers. Distinctive training strategies are also proposed for the generation of effective and diversified ensemble classifiers. The weights of the above ensemble models and the threshold of the novel class confidence are subsequently optimized using a modified Firefly Algorithm, to enhance performance. The above proposed ensemble and optimization algorithms cooperate with each other to conduct the detection of novel unseen classes. Several UCI databases are employed for evaluation, i.e. The KDD Cup, Image Segmentation, Soybean Large, Glass and Iris databases. In comparison with the baseline meta-algorithms such as Boosting, Bagging and Stacking, our approach shows significantly enhanced performance with the increment of the number of novel classes for all the test data sets, which poses great challenges to the existing baseline ensemble methods.
{"title":"Novel Class Detection Using Hybrid Ensemble","authors":"Diptangshu Pandit, Li Zhang, Kamlesh Mistry, Richard M. Jiang","doi":"10.1109/ICMLC51923.2020.9469587","DOIUrl":"https://doi.org/10.1109/ICMLC51923.2020.9469587","url":null,"abstract":"In this research, we propose a hybrid meta-classifier for novel class detection. It is able to efficiently detect the arrival of novel unseen classes as well as tackle real-time data stream classification. Specifically, the proposed hybrid meta-classifler includes three ensemble models, i.e. class-specific, cluster-specific and complementary boosting ensemble classifiers. Distinctive training strategies are also proposed for the generation of effective and diversified ensemble classifiers. The weights of the above ensemble models and the threshold of the novel class confidence are subsequently optimized using a modified Firefly Algorithm, to enhance performance. The above proposed ensemble and optimization algorithms cooperate with each other to conduct the detection of novel unseen classes. Several UCI databases are employed for evaluation, i.e. The KDD Cup, Image Segmentation, Soybean Large, Glass and Iris databases. In comparison with the baseline meta-algorithms such as Boosting, Bagging and Stacking, our approach shows significantly enhanced performance with the increment of the number of novel classes for all the test data sets, which poses great challenges to the existing baseline ensemble methods.","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"190 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133644214","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 : 2020-12-02DOI: 10.1109/ICMLC51923.2020.9469570
Jia-Ching Ying, Chi-Kai Chan, Yen-Ting Chang
At present, the Occupational Accident Labor Insurance premium rate is calculated based on the business categories in Taiwan. The premium rate is calculated as a combination of the experience rate and the manual rate for each business category. The traditional actuarial methods are based on many hypotheses to calculate future actual claims and adjust the rate for each business category. Unfortunately, with such adjustments, the risk level of the insured in the business category will be affected. To accurately estimate the size of actual losses for specific industries, we propose a genetic algorithm applied grouping to determine the premium rate for occupational accidents. The proposed approach has been evaluated using the real-world dataset from the Bureau of Labor Insurance in Taiwan that includes occupational accident insurance data from 2009 to 2015. The results demonstrate that the method is practicable at predicting the applicable premium. The proposed method differs from Taiwan's prevailing occupational accident premium rate calculation method. Moreover, it is efficient at selecting the best group of the Standard Industrial Classification from the genetic algorithm. Lastly, the accuracy of the estimates of the total claim amounts are analyzed.
{"title":"Applying a Genetic Algorithm to Determine Premium Rate of Occupational Accident Insurance","authors":"Jia-Ching Ying, Chi-Kai Chan, Yen-Ting Chang","doi":"10.1109/ICMLC51923.2020.9469570","DOIUrl":"https://doi.org/10.1109/ICMLC51923.2020.9469570","url":null,"abstract":"At present, the Occupational Accident Labor Insurance premium rate is calculated based on the business categories in Taiwan. The premium rate is calculated as a combination of the experience rate and the manual rate for each business category. The traditional actuarial methods are based on many hypotheses to calculate future actual claims and adjust the rate for each business category. Unfortunately, with such adjustments, the risk level of the insured in the business category will be affected. To accurately estimate the size of actual losses for specific industries, we propose a genetic algorithm applied grouping to determine the premium rate for occupational accidents. The proposed approach has been evaluated using the real-world dataset from the Bureau of Labor Insurance in Taiwan that includes occupational accident insurance data from 2009 to 2015. The results demonstrate that the method is practicable at predicting the applicable premium. The proposed method differs from Taiwan's prevailing occupational accident premium rate calculation method. Moreover, it is efficient at selecting the best group of the Standard Industrial Classification from the genetic algorithm. Lastly, the accuracy of the estimates of the total claim amounts are analyzed.","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"50 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124364689","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 : 2020-12-02DOI: 10.1109/ICMLC51923.2020.9469041
N. Takagi, Takashi Suzuki, Tomoyuki Araki
With the development of information processing systems such as screen readers, the accessibility of various information to visually impaired people has been dramatically improved in recent years. In addition, InftyReader, an OCR system for mathematical expression recognition, is making it easier to access scientific documents with mathematical expressions. However, there are still large barriers for blind people to access visual information such as graphs and diagrams. In particular, it is almost impossible for blind people to generate precise figures without the assistance of sighted people. Therefore, we are developing a graphic description language available for the blind. In conventional graphic description languages, it was necessary to accurately specify the coordinates of parameters when drawing elementary shapes. By introducing an object-oriented idea, our language enables users not to specify many coordinates. We have produced an experimental drawing assistant system using our language. In this paper, we outline our language and system, and show the results of experiments which were conducted to verify the effectiveness of our system.
{"title":"Development of a Drawing Assistant System for Blind Users Using an Object-Oriented Graphic Description Language","authors":"N. Takagi, Takashi Suzuki, Tomoyuki Araki","doi":"10.1109/ICMLC51923.2020.9469041","DOIUrl":"https://doi.org/10.1109/ICMLC51923.2020.9469041","url":null,"abstract":"With the development of information processing systems such as screen readers, the accessibility of various information to visually impaired people has been dramatically improved in recent years. In addition, InftyReader, an OCR system for mathematical expression recognition, is making it easier to access scientific documents with mathematical expressions. However, there are still large barriers for blind people to access visual information such as graphs and diagrams. In particular, it is almost impossible for blind people to generate precise figures without the assistance of sighted people. Therefore, we are developing a graphic description language available for the blind. In conventional graphic description languages, it was necessary to accurately specify the coordinates of parameters when drawing elementary shapes. By introducing an object-oriented idea, our language enables users not to specify many coordinates. We have produced an experimental drawing assistant system using our language. In this paper, we outline our language and system, and show the results of experiments which were conducted to verify the effectiveness of our system.","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126138443","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 : 2020-12-02DOI: 10.1109/ICMLC51923.2020.9469534
Ziwen Wang, Haiming Wu, Han Liu, Qianhua Cai
BERT has demonstrated excellent performance in natural language processing due to the training on large amounts of text corpus in an unsupervised way. However, this model is trained to predict the next sentence, and thus it is good at dealing with sentence pair tasks but may not be sufficiently good for other tasks. In our paper, we introduce a novel representation framework BERT-pair-Networks (p-BERTs) for sentiment classification, where p-BERTs involve adopting BERT to encode sentences for sentiment classification as a classic task of single sentence classification, using the auxiliary sentence, and a feature extraction layer on the top. Results on three datasets show that our method achieves considerably improved performance.
{"title":"Bert-Pair-Networks for Sentiment Classification","authors":"Ziwen Wang, Haiming Wu, Han Liu, Qianhua Cai","doi":"10.1109/ICMLC51923.2020.9469534","DOIUrl":"https://doi.org/10.1109/ICMLC51923.2020.9469534","url":null,"abstract":"BERT has demonstrated excellent performance in natural language processing due to the training on large amounts of text corpus in an unsupervised way. However, this model is trained to predict the next sentence, and thus it is good at dealing with sentence pair tasks but may not be sufficiently good for other tasks. In our paper, we introduce a novel representation framework BERT-pair-Networks (p-BERTs) for sentiment classification, where p-BERTs involve adopting BERT to encode sentences for sentiment classification as a classic task of single sentence classification, using the auxiliary sentence, and a feature extraction layer on the top. Results on three datasets show that our method achieves considerably improved performance.","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"452 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121833253","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 : 2020-12-02DOI: 10.1109/ICMLC51923.2020.9469533
C. Fahn, Yi-Lun Wang, Chu-Ping Lee
This paper presents a novel machine reading comprehension model based on deep learning techniques in Chinese environment. In our manner, the training process can be performed using a general-level GPU, and the convergence of the training process can be accelerated for a shorter period of time. In the architectural design, two main constituting parts include Self-Attention Mechanism and Convolutional Neural Networks. To enhance the interaction between an article and questions, we carry out the operation of Context-Query Attention twice, so that our model is more effectively for acquiring the information of the questions related to the article and converges faster while training. In the experiment, we adopt the Delta Reading Comprehension Dataset for model evaluation in Chinese environment. The experimental results reveal that our model is able to reach the accuracy of 64.9% for EM and 79.0% for Fl. The convergence time is less than 1 hour using the Titan XP GPU, and the memory usage is comparatively lower. The training performance is about 3 times faster than other models with state- of-the-art architecture.
提出了一种基于深度学习技术的中文环境下机器阅读理解模型。在我们的方法中，训练过程可以使用通用级GPU来执行，并且可以在更短的时间内加速训练过程的收敛。在架构设计中，自注意机制和卷积神经网络是两个主要组成部分。为了增强文章与问题之间的交互性，我们进行了两次上下文查询关注操作，使我们的模型能够更有效地获取文章相关问题的信息，并且在训练时收敛速度更快。在实验中，我们采用Delta阅读理解数据集进行中文环境下的模型评价。实验结果表明，我们的模型在EM和Fl上的准确率分别达到64.9%和79.0%，在Titan XP GPU上的收敛时间小于1小时，并且内存占用相对较低。训练性能比其他具有最先进架构的模型快3倍左右。
{"title":"A Novel Chinese Reading Comprehension Model Based on Attention Mechanism and Convolutional Neural Networks","authors":"C. Fahn, Yi-Lun Wang, Chu-Ping Lee","doi":"10.1109/ICMLC51923.2020.9469533","DOIUrl":"https://doi.org/10.1109/ICMLC51923.2020.9469533","url":null,"abstract":"This paper presents a novel machine reading comprehension model based on deep learning techniques in Chinese environment. In our manner, the training process can be performed using a general-level GPU, and the convergence of the training process can be accelerated for a shorter period of time. In the architectural design, two main constituting parts include Self-Attention Mechanism and Convolutional Neural Networks. To enhance the interaction between an article and questions, we carry out the operation of Context-Query Attention twice, so that our model is more effectively for acquiring the information of the questions related to the article and converges faster while training. In the experiment, we adopt the Delta Reading Comprehension Dataset for model evaluation in Chinese environment. The experimental results reveal that our model is able to reach the accuracy of 64.9% for EM and 79.0% for Fl. The convergence time is less than 1 hour using the Titan XP GPU, and the memory usage is comparatively lower. The training performance is about 3 times faster than other models with state- of-the-art architecture.","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129317782","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 : 2020-12-02DOI: 10.1109/ICMLC51923.2020.9469586
Naigeng Chen, Chenming Li
Hyperspectral image classification is an important research direction in the application of remote sensing technology. In the process of labeling different types of objects based on spectral information and geometric spatial characteristics, noise interference often exists in continuous multi-band spectral information, which brings great troubles to spectral feature extraction. Besides, far from enough spectral samples will restrict the classification performance of the algorithm to some extent. In order to solve the problem of small amount of original spectral sample data and noisy signal, Wasserstein generative adversarial networks (WGAN) is used to generate samples similar to the original spectrum, and spectral features are extracted from the samples. In the case of small samples, the original materials are provided for the classification of hyperspectral images and a semi-supervised classification model WGAN-CNN for hyperspectral images based on Wasserstein generation antagonistic network is proposed in this paper. This model combines with CNN classifier and completes the classification of terrain objects according to the label for the synthesized samples. The proposed method is compared with several classical hyperspectral image classification methods in classification accuracy. WGAN-CNN can achieve higher classification accuracy in the case of small sample size, which proves the effectiveness of the proposed method.
{"title":"Hyperspectral Image Classification Approach Based on Wasserstein Generative Adversarial Networks","authors":"Naigeng Chen, Chenming Li","doi":"10.1109/ICMLC51923.2020.9469586","DOIUrl":"https://doi.org/10.1109/ICMLC51923.2020.9469586","url":null,"abstract":"Hyperspectral image classification is an important research direction in the application of remote sensing technology. In the process of labeling different types of objects based on spectral information and geometric spatial characteristics, noise interference often exists in continuous multi-band spectral information, which brings great troubles to spectral feature extraction. Besides, far from enough spectral samples will restrict the classification performance of the algorithm to some extent. In order to solve the problem of small amount of original spectral sample data and noisy signal, Wasserstein generative adversarial networks (WGAN) is used to generate samples similar to the original spectrum, and spectral features are extracted from the samples. In the case of small samples, the original materials are provided for the classification of hyperspectral images and a semi-supervised classification model WGAN-CNN for hyperspectral images based on Wasserstein generation antagonistic network is proposed in this paper. This model combines with CNN classifier and completes the classification of terrain objects according to the label for the synthesized samples. The proposed method is compared with several classical hyperspectral image classification methods in classification accuracy. WGAN-CNN can achieve higher classification accuracy in the case of small sample size, which proves the effectiveness of the proposed method.","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"302 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132096272","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 : 2020-12-02DOI: 10.1109/ICMLC51923.2020.9469043
Zonglin Yang, Qiang Zhao
Pneumonia is a lung disease caused by bacterial or viral infection. Early diagnosis is an important factor for successful treatment. In this study, we use three well-known convolutional neural network models, namely Faster RCNN ResNet-101, Mask RCNN ResNet-101, and Mask RCNN ResNet-50 for detection of pneumonia. We use data augmentation, transfer learning and fine-tuning in the training stage. Experimental results show that different networks have different characteristics on the same dataset. Therefore, we propose a multiple deep learner approach to improve the prediction performance via combination of different object detection models. As a result, the proposed approach can find more opacity areas of the lungs where the early symptoms are not evident. While maintaining the prediction accuracy, the proposed method can predict the bounding box size more precisely with a higher confidence score.
{"title":"A Multiple Deep Learner Approach for X-Ray Image-Based Pneumonia Detection","authors":"Zonglin Yang, Qiang Zhao","doi":"10.1109/ICMLC51923.2020.9469043","DOIUrl":"https://doi.org/10.1109/ICMLC51923.2020.9469043","url":null,"abstract":"Pneumonia is a lung disease caused by bacterial or viral infection. Early diagnosis is an important factor for successful treatment. In this study, we use three well-known convolutional neural network models, namely Faster RCNN ResNet-101, Mask RCNN ResNet-101, and Mask RCNN ResNet-50 for detection of pneumonia. We use data augmentation, transfer learning and fine-tuning in the training stage. Experimental results show that different networks have different characteristics on the same dataset. Therefore, we propose a multiple deep learner approach to improve the prediction performance via combination of different object detection models. As a result, the proposed approach can find more opacity areas of the lungs where the early symptoms are not evident. While maintaining the prediction accuracy, the proposed method can predict the bounding box size more precisely with a higher confidence score.","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"56 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127824933","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 : 2020-12-02DOI: 10.1109/icmlc51923.2020.9469563
{"title":"[Copyright notice]","authors":"","doi":"10.1109/icmlc51923.2020.9469563","DOIUrl":"https://doi.org/10.1109/icmlc51923.2020.9469563","url":null,"abstract":"","PeriodicalId":170815,"journal":{"name":"2020 International Conference on Machine Learning and Cybernetics (ICMLC)","volume":"58 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128218982","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：