Pub Date : 2020-03-01DOI: 10.1109/ESCI48226.2020.9167616
Mohammad Alamgir Hossain, Basem Assiri
Recognition and classification of face-emotion is a vital issue now a day. Emotion bears a resemblance to the people's thought process and based on that a mapping of anyone's activity is possible to establish by analyzing facial expressions. Facial emotion is recognized based on interaction or appearances of eyes, chick, forehead, lips as well as from the whole face in different forms. In this paper, facial emotion is recognized and classified them to create infrared thermal face image data-mask and tried to correlate them based on the variances and standard deviation with EPDF (Enhanced Probability Density Function) of the identified images. During the testing and recognition process, a centralized stored data has been is used to avoid redundancy of data to be stored after recognition. In this experiment, three types of emotions are taken into account and their infrared thermal facial images are recorded simultaneously. In the processing, a calibration procedure is adopted to reduce the variances produced by dissimilar image-set from the same face due to independent parts of a face analysis that are related to facial emotions. Features are taken out from pixel values of classified images. The investigational results of facial images confirmed that the proposed system attained 91.73% accuracy in identification in RGB and 92.39% in infrared images respectively. Whereas as per D. Kumar et al, it is 65% and M. A. Eid has achieved 85% accuracy on identification. The average detection is 91.73% with eight RGB images. Whereas detection from the eight infrared images the average detection rate is 92.39%. This exits the robustness of the suggested methods.
面部表情的识别和分类是当今社会的一个重要问题。情感与人的思维过程有相似之处，在此基础上，通过分析面部表情可以绘制出任何人的活动图谱。面部情绪的识别是基于眼睛、小鸡、额头、嘴唇的相互作用或外观，以及整个面部的不同形式。本文对人脸情绪进行识别和分类，创建红外热人脸图像数据掩模，并尝试根据识别图像的方差和标准差与EPDF (Enhanced Probability Density Function)进行关联。在测试和识别过程中，采用了集中存储数据的方式，避免了识别后存储数据的冗余。在本实验中，考虑了三种类型的情绪，同时记录了它们的红外热面部图像。在处理过程中，采用了一种校准程序，以减少由于人脸分析中与面部情绪相关的独立部分而导致的来自同一人脸的不同图像集产生的方差。从分类图像的像素值中提取特征。人脸图像的研究结果表明，该系统在RGB图像和红外图像上的识别准确率分别达到91.73%和92.39%。而根据D. Kumar等人的研究，识别准确率为65%，m.a. Eid的识别准确率达到85%。8张RGB图像的平均检出率为91.73%。而对8幅红外图像的平均检出率为92.39%。这使得所建议的方法具有鲁棒性。
{"title":"Facial Emotion Verification by Infrared Image","authors":"Mohammad Alamgir Hossain, Basem Assiri","doi":"10.1109/ESCI48226.2020.9167616","DOIUrl":"https://doi.org/10.1109/ESCI48226.2020.9167616","url":null,"abstract":"Recognition and classification of face-emotion is a vital issue now a day. Emotion bears a resemblance to the people's thought process and based on that a mapping of anyone's activity is possible to establish by analyzing facial expressions. Facial emotion is recognized based on interaction or appearances of eyes, chick, forehead, lips as well as from the whole face in different forms. In this paper, facial emotion is recognized and classified them to create infrared thermal face image data-mask and tried to correlate them based on the variances and standard deviation with EPDF (Enhanced Probability Density Function) of the identified images. During the testing and recognition process, a centralized stored data has been is used to avoid redundancy of data to be stored after recognition. In this experiment, three types of emotions are taken into account and their infrared thermal facial images are recorded simultaneously. In the processing, a calibration procedure is adopted to reduce the variances produced by dissimilar image-set from the same face due to independent parts of a face analysis that are related to facial emotions. Features are taken out from pixel values of classified images. The investigational results of facial images confirmed that the proposed system attained 91.73% accuracy in identification in RGB and 92.39% in infrared images respectively. Whereas as per D. Kumar et al, it is 65% and M. A. Eid has achieved 85% accuracy on identification. The average detection is 91.73% with eight RGB images. Whereas detection from the eight infrared images the average detection rate is 92.39%. This exits the robustness of the suggested methods.","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114664064","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 : 2020-03-01DOI: 10.1109/ESCI48226.2020.9167632
P. Bafna, Jatinderkumar R. Saini
Document management is an essential but critical task. Categorizing these documents into the groups benefits many applications in commercial, industrial and other domains. Manual efforts are reduced by placing documents into its corresponding class. And predicting the category of document. It also reduces the time which otherwise would have required to read the document. Hindi has gained significant value in different fields like information technology, since the last decade due to the multilinguistic talent supported by websites. Natural Language toolkits along with text mining generate speedy, economic and scalable results. In spite of gaining importance in the digital era, Hindi document classification is targeted by very few researchers. Two eager machine learning algorithms are applied on the corpus containing 450 Hindi poems. Poetry/poem gets classified based on terms present in it. The classifiers are evaluated using a misclassification error.
{"title":"Hindi Poetry Classification using Eager Supervised Machine Learning Algorithms","authors":"P. Bafna, Jatinderkumar R. Saini","doi":"10.1109/ESCI48226.2020.9167632","DOIUrl":"https://doi.org/10.1109/ESCI48226.2020.9167632","url":null,"abstract":"Document management is an essential but critical task. Categorizing these documents into the groups benefits many applications in commercial, industrial and other domains. Manual efforts are reduced by placing documents into its corresponding class. And predicting the category of document. It also reduces the time which otherwise would have required to read the document. Hindi has gained significant value in different fields like information technology, since the last decade due to the multilinguistic talent supported by websites. Natural Language toolkits along with text mining generate speedy, economic and scalable results. In spite of gaining importance in the digital era, Hindi document classification is targeted by very few researchers. Two eager machine learning algorithms are applied on the corpus containing 450 Hindi poems. Poetry/poem gets classified based on terms present in it. The classifiers are evaluated using a misclassification error.","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127453855","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 : 2020-03-01DOI: 10.1109/ESCI48226.2020.9167646
Prashant U. Jain, V. Tomar
In the era of smart computing, almost 85-90% area is captured by memories in digital designs. In order to reduce the power dissipation and improve the overall performance of digital logic circuits, conventional MOSFET technology may replace by FinFET technology. FinFETs are the best choice as an alternative for MOSFET below 32nm technology, as below 32nm short channel effects may introduce more problems. With low leakage and low power feature, FinFET technology becomes very popular and widely used instead of conventional MOS almost in all digital circuits. In this paper, FinFET technology has been demonstrated as a good alternative of conventional CMOS technology.
{"title":"FinFET Technology : As A Promising Alternatives for Conventional MOSFET Technology","authors":"Prashant U. Jain, V. Tomar","doi":"10.1109/ESCI48226.2020.9167646","DOIUrl":"https://doi.org/10.1109/ESCI48226.2020.9167646","url":null,"abstract":"In the era of smart computing, almost 85-90% area is captured by memories in digital designs. In order to reduce the power dissipation and improve the overall performance of digital logic circuits, conventional MOSFET technology may replace by FinFET technology. FinFETs are the best choice as an alternative for MOSFET below 32nm technology, as below 32nm short channel effects may introduce more problems. With low leakage and low power feature, FinFET technology becomes very popular and widely used instead of conventional MOS almost in all digital circuits. In this paper, FinFET technology has been demonstrated as a good alternative of conventional CMOS technology.","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130324525","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 : 2020-03-01DOI: 10.1109/ESCI48226.2020.9167568
Tawseef Ayoub Shaikh, Rashid Ali
Even though today's medical field is technologically advanced, some diseases still haunt the human race by posing a hustle in its existence. In addition to the sophisticated tools and techniques for disease diagnosis, recent use of information and communication technology (ICT) has also tightened its spine to serve this noble cause. To have an interior view of the body without surgery, medical imaging is a predominant technique behind early/automatic diagnosing and detecting diseases. Image mammography is the primary asset assisting doctors for having a projection of the interiors of breast tissues, thus offering a crucial big stick in the diagnosis of malignancy/non- malignancy in the tissues. Using certain supervised and unsupervised filters offered by Weka on top of BCDR-F01 cancer benhcmark dataset, this work intends to increase the objectivity of clinical diagnostics. Misclassification cost of Naive Bayes algorithm is measured and compared the same with misclassification cost measured after applying respective filters. The results show the accuracies in case of supervised attribute DISCRETIZE filter, supervised instance RESAMPLE filter and unsupervised attribute PKIIDiscretize filter get amplified from 73.7569 % to 81.768 %, 85.0829 %, and 78.7293 % and only in case of unsupervised instance RESAMPLE filter, it shows a minute decrease to 73.4807 %.
{"title":"A CAD Tool for Breast Cancer Prediction using Naive Bayes Classifier","authors":"Tawseef Ayoub Shaikh, Rashid Ali","doi":"10.1109/ESCI48226.2020.9167568","DOIUrl":"https://doi.org/10.1109/ESCI48226.2020.9167568","url":null,"abstract":"Even though today's medical field is technologically advanced, some diseases still haunt the human race by posing a hustle in its existence. In addition to the sophisticated tools and techniques for disease diagnosis, recent use of information and communication technology (ICT) has also tightened its spine to serve this noble cause. To have an interior view of the body without surgery, medical imaging is a predominant technique behind early/automatic diagnosing and detecting diseases. Image mammography is the primary asset assisting doctors for having a projection of the interiors of breast tissues, thus offering a crucial big stick in the diagnosis of malignancy/non- malignancy in the tissues. Using certain supervised and unsupervised filters offered by Weka on top of BCDR-F01 cancer benhcmark dataset, this work intends to increase the objectivity of clinical diagnostics. Misclassification cost of Naive Bayes algorithm is measured and compared the same with misclassification cost measured after applying respective filters. The results show the accuracies in case of supervised attribute DISCRETIZE filter, supervised instance RESAMPLE filter and unsupervised attribute PKIIDiscretize filter get amplified from 73.7569 % to 81.768 %, 85.0829 %, and 78.7293 % and only in case of unsupervised instance RESAMPLE filter, it shows a minute decrease to 73.4807 %.","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"10 11","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114030508","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 : 2020-03-01DOI: 10.1109/ESCI48226.2020.9167512
S. S. Sengar, Hariharan U, K. Rajkumar
For specific identification process, Identity Management details an ailment of supplying authorized owners with secure and easy admittance to information and solutions. For choosing the individual's identity, the primary goal is actually executing secured identification feature. PINs, keys, gain access to cards, passwords, tokens are actually the private determining elements which are actually utilized within standard methods which may have a tendency to drawbacks such as cracking, stealing, copying and posting. Biometrics grounded identification is needed having a perspective to stay away from the drawbacks. Due to intra category variants, non- universality, sound as well as spoof strikes are impacted. Multimodal biometrics are actually employed to get rid of the episodes which are actually a grouping of countless modalities. For an authentication supply, Fingerprint and Palmprint identification are popular systems these days. For minutiae thing detection as well as attribute extraction, with this paper, rich neural community (DNN) were definitely projected. The confinements of unimodal biometric structure lead to substantial False Acceptance Rate (FAR) along with False Rejection Rate (FRR), limited splitting up skill, top bound within delivery therefore the multimodal biometric product is designed to satisfy the strict delivery demands. For minutiae corresponding, values of Euclidean distance are actually used. The better identification pace is actually attained throughout the suggested procedure & it's extremely safe only in loud problem.
{"title":"Multimodal Biometric Authentication System using Deep Learning Method","authors":"S. S. Sengar, Hariharan U, K. Rajkumar","doi":"10.1109/ESCI48226.2020.9167512","DOIUrl":"https://doi.org/10.1109/ESCI48226.2020.9167512","url":null,"abstract":"For specific identification process, Identity Management details an ailment of supplying authorized owners with secure and easy admittance to information and solutions. For choosing the individual's identity, the primary goal is actually executing secured identification feature. PINs, keys, gain access to cards, passwords, tokens are actually the private determining elements which are actually utilized within standard methods which may have a tendency to drawbacks such as cracking, stealing, copying and posting. Biometrics grounded identification is needed having a perspective to stay away from the drawbacks. Due to intra category variants, non- universality, sound as well as spoof strikes are impacted. Multimodal biometrics are actually employed to get rid of the episodes which are actually a grouping of countless modalities. For an authentication supply, Fingerprint and Palmprint identification are popular systems these days. For minutiae thing detection as well as attribute extraction, with this paper, rich neural community (DNN) were definitely projected. The confinements of unimodal biometric structure lead to substantial False Acceptance Rate (FAR) along with False Rejection Rate (FRR), limited splitting up skill, top bound within delivery therefore the multimodal biometric product is designed to satisfy the strict delivery demands. For minutiae corresponding, values of Euclidean distance are actually used. The better identification pace is actually attained throughout the suggested procedure & it's extremely safe only in loud problem.","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134264888","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 : 2020-03-01DOI: 10.1109/ESCI48226.2020.9167537
M. Rani, A. Bakshi, Akhil Gupta
These days, wellbeing disease are expanding step by step as a result of life vogue, inherited. Particularly, cardiopathy has turned into a great deal of basic as of late .for example lifetime of people is in peril. Each individual has totally extraordinary qualities for power per unit region, cholesterol and essential sign. Anyway per restoratively attempted outcomes the customary estimations of power per unit territory is 120/90, cholesterol is and essential sign is seventy two. This paper gives the study with respect to totally extraordinary arrangement systems utilized for anticipating the opportunity dimension of each individual bolstered age, sexual orientation, constrain per unit zone, cholesterol, beat rate. We will utilize naïve bayes and image processing to predict the heart disease efficiently.
{"title":"Prediction of Heart Disease Using Naïve bayes and Image Processing","authors":"M. Rani, A. Bakshi, Akhil Gupta","doi":"10.1109/ESCI48226.2020.9167537","DOIUrl":"https://doi.org/10.1109/ESCI48226.2020.9167537","url":null,"abstract":"These days, wellbeing disease are expanding step by step as a result of life vogue, inherited. Particularly, cardiopathy has turned into a great deal of basic as of late .for example lifetime of people is in peril. Each individual has totally extraordinary qualities for power per unit region, cholesterol and essential sign. Anyway per restoratively attempted outcomes the customary estimations of power per unit territory is 120/90, cholesterol is and essential sign is seventy two. This paper gives the study with respect to totally extraordinary arrangement systems utilized for anticipating the opportunity dimension of each individual bolstered age, sexual orientation, constrain per unit zone, cholesterol, beat rate. We will utilize naïve bayes and image processing to predict the heart disease efficiently.","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131807332","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 : 2020-03-01DOI: 10.1109/ESCI48226.2020.9167513
Surbhi Sharma, B. Kaushik
Internet of Vehicles is an integration of VANETs and IoT to enhance the proficiency of VANETs by incorporating smartness. Due to its numerous characteristics, it has gained lot of attention among researchers. Nature inspired algorithms are inspired from nature's strategy to cope with all day to day problems. In this review, main focus is to explore nature-inspired algorithms as these are quite beneficial in optimization. Nature-inspired algorithms are capable to deal with all kind of complex problems so in this paper, its applicability in internet of vehicles has been explored. In internet of vehicles, nature –inspired algorithms can be applied mainly in two aspects-Routing and Security. It aims to optimize all routing issues among vehicles as delay and timely information cannot be tolerated in real-time applications. On the other hand, security is of major concern in vehicular networks thus, nature inspired algorithms are able to prevent various attacks. In this paper, we have reviewed both routing and security applications of nature-inspired algorithms.
{"title":"A Comprehensive Review of Nature-inspired Algorithms for Internet of Vehicles","authors":"Surbhi Sharma, B. Kaushik","doi":"10.1109/ESCI48226.2020.9167513","DOIUrl":"https://doi.org/10.1109/ESCI48226.2020.9167513","url":null,"abstract":"Internet of Vehicles is an integration of VANETs and IoT to enhance the proficiency of VANETs by incorporating smartness. Due to its numerous characteristics, it has gained lot of attention among researchers. Nature inspired algorithms are inspired from nature's strategy to cope with all day to day problems. In this review, main focus is to explore nature-inspired algorithms as these are quite beneficial in optimization. Nature-inspired algorithms are capable to deal with all kind of complex problems so in this paper, its applicability in internet of vehicles has been explored. In internet of vehicles, nature –inspired algorithms can be applied mainly in two aspects-Routing and Security. It aims to optimize all routing issues among vehicles as delay and timely information cannot be tolerated in real-time applications. On the other hand, security is of major concern in vehicular networks thus, nature inspired algorithms are able to prevent various attacks. In this paper, we have reviewed both routing and security applications of nature-inspired algorithms.","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131082719","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 : 2020-03-01DOI: 10.1109/ESCI48226.2020.9167649
Reya Sharma, B. Kaushik, N. Gondhi
Digitization of machine printed or handwritten text documents have become very popular with the advancements in computing and technology. Humans have tried to automatized their work by replacing themselves with machines. The transformation from manual to automatization gave rise to several research areas and text recognition is one among them. Deep learning and machine learning techniques have been proved to be very suitable for optical character recognition. In this work, an up-to-date overview of four machine learning and deep learning architectures, viz., Support vector machine, Artificial neural network, Naive Bayes and Convolutional neural network have been discussed in detail.
{"title":"Character Recognition using Machine Learning and Deep Learning - A Survey","authors":"Reya Sharma, B. Kaushik, N. Gondhi","doi":"10.1109/ESCI48226.2020.9167649","DOIUrl":"https://doi.org/10.1109/ESCI48226.2020.9167649","url":null,"abstract":"Digitization of machine printed or handwritten text documents have become very popular with the advancements in computing and technology. Humans have tried to automatized their work by replacing themselves with machines. The transformation from manual to automatization gave rise to several research areas and text recognition is one among them. Deep learning and machine learning techniques have been proved to be very suitable for optical character recognition. In this work, an up-to-date overview of four machine learning and deep learning architectures, viz., Support vector machine, Artificial neural network, Naive Bayes and Convolutional neural network have been discussed in detail.","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129732556","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 : 2020-03-01DOI: 10.1109/ESCI48226.2020.9167650
Pangambam Sendash Singh, V. Singh, M. Pandey, S. Karthikeyan
The scarcity of labelled training data as well as uneven class distribution among the limitedly available labelled data have posed a critical issue in supervised hyperspectral remote sensing image classification. Semisupervised methods can be an easy solution to this critical problem. However, traditional self-training based semi-supervised approaches often give poor classification results in high dimensional multiclass classification problems. This paper proposes a novel efficient one-class classifier ensemble based self-training approach for semisupervised classification of hyperspectral remote sensing images with limited labelled data. The proposed method initially trains an ensemble of locally specialized one-class classifiers independently by using the dimensionally reduced spectral feature vectors of the available labelled samples. The trained one-class classifiers are then used to extend the labelled set by iterative addition of high quality unlabelled samples to it through the exploitation of both spectral and spatial information. The classifiers are then retrained with the extended dataset in a batchwise fashion. The procedure is repeated until an adequate quantity of labelled samples are generated. Finally, a supervised multiclass classifier is trained on the extended dataset for the final image classification purpose. Experimental results on two benchmark hyperspectral images verify the effectiveness of the proposed method over supervised and traditional self-training based semisupervised pixelwise classification in terms of different classification measures.
{"title":"One-class Classifier Ensemble based Enhanced Semisupervised Classification of Hyperspectral Remote Sensing Images","authors":"Pangambam Sendash Singh, V. Singh, M. Pandey, S. Karthikeyan","doi":"10.1109/ESCI48226.2020.9167650","DOIUrl":"https://doi.org/10.1109/ESCI48226.2020.9167650","url":null,"abstract":"The scarcity of labelled training data as well as uneven class distribution among the limitedly available labelled data have posed a critical issue in supervised hyperspectral remote sensing image classification. Semisupervised methods can be an easy solution to this critical problem. However, traditional self-training based semi-supervised approaches often give poor classification results in high dimensional multiclass classification problems. This paper proposes a novel efficient one-class classifier ensemble based self-training approach for semisupervised classification of hyperspectral remote sensing images with limited labelled data. The proposed method initially trains an ensemble of locally specialized one-class classifiers independently by using the dimensionally reduced spectral feature vectors of the available labelled samples. The trained one-class classifiers are then used to extend the labelled set by iterative addition of high quality unlabelled samples to it through the exploitation of both spectral and spatial information. The classifiers are then retrained with the extended dataset in a batchwise fashion. The procedure is repeated until an adequate quantity of labelled samples are generated. Finally, a supervised multiclass classifier is trained on the extended dataset for the final image classification purpose. Experimental results on two benchmark hyperspectral images verify the effectiveness of the proposed method over supervised and traditional self-training based semisupervised pixelwise classification in terms of different classification measures.","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131981673","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 : 2020-03-01DOI: 10.1109/esci48226.2020.9167658
{"title":"Our Heritage","authors":"","doi":"10.1109/esci48226.2020.9167658","DOIUrl":"https://doi.org/10.1109/esci48226.2020.9167658","url":null,"abstract":"","PeriodicalId":401691,"journal":{"name":"2020 International Conference on Emerging Smart Computing and Informatics (ESCI)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128825945","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：