Pub Date : 2018-02-01DOI: 10.1109/SPIN.2018.8474035
Vasudha, D. Kakkar
In this paper, local directional position pattern (LDPP) and local ternary pattern (LTP) are selected for facial recognition method which are having many advantages over previous techniques like local binary pattern (LBP) and local directional pattern (LDP). The selected techniques of LDPP and LTP are estrangement in their algorithms which help solely to extract features out of an image. LDPP is a revised form of LDP. In a typical LDP, only the top edge direction was taken into consideration, but strength sign of the pixel was not considered which may result in same code for opposite kind of edge pixel. This snag is overcome by LDPP which is further concatenated with LTP for better feature extraction. Once features are extracted they are trained using deep belief network. In the experimental work 10 images of each expression i.e. angry, surprise, disgust, neutral, sad, smile are selected. LDPP and LTP are concatenated followed by principal component analysis (PCA) and general discriminant analysis (GDA). Further for training, Deep Belief Network (DBN) is used which eventually increases the recognition rate and achieve accuracy of 95.3% which was 89.3% without concatenating.
{"title":"Facial Expression Recognition with LDPP & LTP using Deep Belief Network","authors":"Vasudha, D. Kakkar","doi":"10.1109/SPIN.2018.8474035","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474035","url":null,"abstract":"In this paper, local directional position pattern (LDPP) and local ternary pattern (LTP) are selected for facial recognition method which are having many advantages over previous techniques like local binary pattern (LBP) and local directional pattern (LDP). The selected techniques of LDPP and LTP are estrangement in their algorithms which help solely to extract features out of an image. LDPP is a revised form of LDP. In a typical LDP, only the top edge direction was taken into consideration, but strength sign of the pixel was not considered which may result in same code for opposite kind of edge pixel. This snag is overcome by LDPP which is further concatenated with LTP for better feature extraction. Once features are extracted they are trained using deep belief network. In the experimental work 10 images of each expression i.e. angry, surprise, disgust, neutral, sad, smile are selected. LDPP and LTP are concatenated followed by principal component analysis (PCA) and general discriminant analysis (GDA). Further for training, Deep Belief Network (DBN) is used which eventually increases the recognition rate and achieve accuracy of 95.3% which was 89.3% without concatenating.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"81 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116045687","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 : 2018-02-01DOI: 10.1109/SPIN.2018.8474170
Sachin Doshi, Malvi Shrimali, S. K. Rajendra, M. Sharma
Mechanical deformations in the power transformer are the result of short circuit forces and improper handling of transformer during transportation. Such deformations grow with the time and might lead to complete breakdown of the transformer. Hence, monitoring the condition of the transformer is essential. This paper presents a technique to analyse the terminal behaviour of the transformer winding. To this end, high frequency circuit model of the transformer winding comprises of inductances, capacitances and resistances is considered initially. Mechanical deformations are then introduced by changing these circuit parameters. Frequency response analysis (FRA) is performed to obtain terminal behavior of the circuit model under both healthy and unhealthy conditions. Signals obtained from FRA are decomposed into five subbands (SBs) using tunable Q-Factor wavelet transform (TQWT). Afterwards, with the help of Shannon Entropy (SE), features of the SBs are extracted. These features are then classified using K-nearest neighbour (KNN) and Ensemble Bagged algorithm (EB). The statistical parameters like p-value and t-test clearly indicated that the signals are classified in to healthy and faulty states. Further, data have been classified properly with an accuracy of 99.2%.
{"title":"Tunable Q-Factor Wavelet Transform for Classifying Mechanical Deformations in Power Transformer","authors":"Sachin Doshi, Malvi Shrimali, S. K. Rajendra, M. Sharma","doi":"10.1109/SPIN.2018.8474170","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474170","url":null,"abstract":"Mechanical deformations in the power transformer are the result of short circuit forces and improper handling of transformer during transportation. Such deformations grow with the time and might lead to complete breakdown of the transformer. Hence, monitoring the condition of the transformer is essential. This paper presents a technique to analyse the terminal behaviour of the transformer winding. To this end, high frequency circuit model of the transformer winding comprises of inductances, capacitances and resistances is considered initially. Mechanical deformations are then introduced by changing these circuit parameters. Frequency response analysis (FRA) is performed to obtain terminal behavior of the circuit model under both healthy and unhealthy conditions. Signals obtained from FRA are decomposed into five subbands (SBs) using tunable Q-Factor wavelet transform (TQWT). Afterwards, with the help of Shannon Entropy (SE), features of the SBs are extracted. These features are then classified using K-nearest neighbour (KNN) and Ensemble Bagged algorithm (EB). The statistical parameters like p-value and t-test clearly indicated that the signals are classified in to healthy and faulty states. Further, data have been classified properly with an accuracy of 99.2%.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115965205","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 : 2018-02-01DOI: 10.1109/SPIN.2018.8474042
Monika Mathur, Nidhi Goel
Absorption and scattering are the two main issues in underwater image capturing which results in diminished contrast, high level of blurring and faded colours. The present paper proposes a method of enhancing underwater images for better visual quality using auto white balancing followed by gamma correction and Rayleigh stretching in RGB colour model. The proposed method is very effective which does not require any dedicated hardware and depends only on single image. The use of white balancing compensates the non-uniform color cast which is caused by the selective absorption of colors with depth. Histogram stretching of red colour channel is done with a minimum limit of 5% and histogram stretching of blue channel is done with a maximum limit of 95%. Histogram of green channel is stretched in both the directions. To compensate the effect of under and over enhancement from the image, histograms of the stretched colour channels are mapped to trail a Rayleigh distribution. Thorough analysis has been carried out to verify the proposed method. Contrast and image details are effectively enhanced by minimizing the bluish-green effect and reducing less as well as more enhanced areas from resultant image.
{"title":"Enhancement of Underwater images using White Balancing and Rayleigh-Stretching","authors":"Monika Mathur, Nidhi Goel","doi":"10.1109/SPIN.2018.8474042","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474042","url":null,"abstract":"Absorption and scattering are the two main issues in underwater image capturing which results in diminished contrast, high level of blurring and faded colours. The present paper proposes a method of enhancing underwater images for better visual quality using auto white balancing followed by gamma correction and Rayleigh stretching in RGB colour model. The proposed method is very effective which does not require any dedicated hardware and depends only on single image. The use of white balancing compensates the non-uniform color cast which is caused by the selective absorption of colors with depth. Histogram stretching of red colour channel is done with a minimum limit of 5% and histogram stretching of blue channel is done with a maximum limit of 95%. Histogram of green channel is stretched in both the directions. To compensate the effect of under and over enhancement from the image, histograms of the stretched colour channels are mapped to trail a Rayleigh distribution. Thorough analysis has been carried out to verify the proposed method. Contrast and image details are effectively enhanced by minimizing the bluish-green effect and reducing less as well as more enhanced areas from resultant image.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"81 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127725747","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 : 2018-02-01DOI: 10.1109/SPIN.2018.8474205
Anamika Baishya, Priyatam Kumar
This paper presents an improved speech enhancement technique based on wavelet transform along with excitation-based classification of speech to eliminate noise from speech signals. The method initially classifies the speech into voiced, unvoiced and silence regions on the basis of a novel energy-based threshold and then wavelet transform is applied. To remove the noise, thresholding is applied to the detail coefficients by taking into consideration different characteristics of speech in the three different regions. For this, soft thresholding is used for the voiced regions, hard thresholding for the unvoiced regions and the wavelet coefficients of silence regions are made zero. Speech signals obtained from SPEAR database and corrupted with white noise are being used for evaluation of the proposed method. Experimental results show, in terms of SNR and PESQ score, de-noising of speech is achieved using the proposed method. With regards to SNR, the best improvement is 9.4 dB when compared to the SNR of the original (noisy) speech and 1.2 dB as compared to the improvement obtained using one of the recently reported methods.
{"title":"Speech De-noising using Wavelet based Methods with Focus on Classification of Speech into Voiced, Unvoiced and Silence Regions","authors":"Anamika Baishya, Priyatam Kumar","doi":"10.1109/SPIN.2018.8474205","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474205","url":null,"abstract":"This paper presents an improved speech enhancement technique based on wavelet transform along with excitation-based classification of speech to eliminate noise from speech signals. The method initially classifies the speech into voiced, unvoiced and silence regions on the basis of a novel energy-based threshold and then wavelet transform is applied. To remove the noise, thresholding is applied to the detail coefficients by taking into consideration different characteristics of speech in the three different regions. For this, soft thresholding is used for the voiced regions, hard thresholding for the unvoiced regions and the wavelet coefficients of silence regions are made zero. Speech signals obtained from SPEAR database and corrupted with white noise are being used for evaluation of the proposed method. Experimental results show, in terms of SNR and PESQ score, de-noising of speech is achieved using the proposed method. With regards to SNR, the best improvement is 9.4 dB when compared to the SNR of the original (noisy) speech and 1.2 dB as compared to the improvement obtained using one of the recently reported methods.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127998301","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 : 2018-02-01DOI: 10.1109/SPIN.2018.8474049
V. Trivedi, Madhusudan Kumar Sinha, Preetam Kumar
Single Carrier Frequency Division Multiple Access (SC-FDMA) is a promising technique compared to Orthogonal Frequency Division Multiple Access (OFDMA) for uplink transmission in 3rd Generation Partnership Project(3GPP) Long Term Evolution(LTE) because of its low peak to average power ratio (PAPR). Being a multiple access technique, resources are required to be distributed among various users. In this paper, we have used Zero Forcing noise amplification factor (βZF) for scheduling subcarrier allocation in SC-FDMA systems unlike using the summation of amplitude response over different subcarriers. Using βZF for scheduling gives more exact information about channel condition over different sub bands for different users. For scheduling purpose we have used, Round Robin, Greedy Chunk and Proportionally Fair algorithms and we have compared the SER results for the different number of users including maximum possible users transmitting simultaneously. The effect on capacity and throughput over error performance has been observed earlier. In this work, the effects of different channel-dependent scheduling schemes on error rate performance of SC-FDMA have been evaluated.
{"title":"Error Rate Performance of SC-FDMA with Channel Dependent Subcarrier Scheduling","authors":"V. Trivedi, Madhusudan Kumar Sinha, Preetam Kumar","doi":"10.1109/SPIN.2018.8474049","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474049","url":null,"abstract":"Single Carrier Frequency Division Multiple Access (SC-FDMA) is a promising technique compared to Orthogonal Frequency Division Multiple Access (OFDMA) for uplink transmission in 3rd Generation Partnership Project(3GPP) Long Term Evolution(LTE) because of its low peak to average power ratio (PAPR). Being a multiple access technique, resources are required to be distributed among various users. In this paper, we have used Zero Forcing noise amplification factor (βZF) for scheduling subcarrier allocation in SC-FDMA systems unlike using the summation of amplitude response over different subcarriers. Using βZF for scheduling gives more exact information about channel condition over different sub bands for different users. For scheduling purpose we have used, Round Robin, Greedy Chunk and Proportionally Fair algorithms and we have compared the SER results for the different number of users including maximum possible users transmitting simultaneously. The effect on capacity and throughput over error performance has been observed earlier. In this work, the effects of different channel-dependent scheduling schemes on error rate performance of SC-FDMA have been evaluated.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"438 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132800417","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 : 2018-02-01DOI: 10.1109/SPIN.2018.8474100
Hemant Choubey, Alpana Pandey
The Electroencephalogram (EEG) signal is the non-invasive technique to examine the electrical activity of the brain and epilepsy is the chronological disorder or abnormality symptoms obtained from EEG data. The detection of this abnormality requires large number of features for the classification of healthy, inter-ictal and ictal signal from the EEG signal. Epileptic seizure detection using reduced set of features is the main idea behind in this paper. Expected Activity Measurement coefficient and Hurst Exponent with Higuchi Fractal Dimension is the small set of features sufficient for the detection of epileptic seizure from EEG signal using k-NN classifier with performance parameter like Accuracy, Precision and Jaccard Coefficient.
{"title":"Classification and Detection of Epilepsy using Reduced Set of Extracted Features","authors":"Hemant Choubey, Alpana Pandey","doi":"10.1109/SPIN.2018.8474100","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474100","url":null,"abstract":"The Electroencephalogram (EEG) signal is the non-invasive technique to examine the electrical activity of the brain and epilepsy is the chronological disorder or abnormality symptoms obtained from EEG data. The detection of this abnormality requires large number of features for the classification of healthy, inter-ictal and ictal signal from the EEG signal. Epileptic seizure detection using reduced set of features is the main idea behind in this paper. Expected Activity Measurement coefficient and Hurst Exponent with Higuchi Fractal Dimension is the small set of features sufficient for the detection of epileptic seizure from EEG signal using k-NN classifier with performance parameter like Accuracy, Precision and Jaccard Coefficient.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"100 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133281596","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 : 2018-02-01DOI: 10.1109/SPIN.2018.8474203
Nirdosh, Shalini Sah, Ashna Kakkar
A two-element wideband MIMO antenna is presented in this letter. This wideband MIMO is composed of two inverted E-shaped monopole elements having symmetrically configuration, and uses of dumbbell shaped slot structure between central positions of two elements to achieve low mutual coupling. It has covered wide frequency bandwidth from 8.47 GHz to 26.76 GHz with higher isolation (more than 22 dB). The overall dimension of this antenna is 35mm × 40mm × 1.6mm. This simulated antenna is realized on FR4 substrate with relative permittivity of 4.4.The ECC (envelope correlation coefficient) and diversity gain are also studied and calculated which provide satisfactory results over the entire operating band for the application of X-band, Ku-band and K-band.
{"title":"A Two-element Wideband MIMO Antenna For X-Band, Ku-Band, K-band Applications","authors":"Nirdosh, Shalini Sah, Ashna Kakkar","doi":"10.1109/SPIN.2018.8474203","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474203","url":null,"abstract":"A two-element wideband MIMO antenna is presented in this letter. This wideband MIMO is composed of two inverted E-shaped monopole elements having symmetrically configuration, and uses of dumbbell shaped slot structure between central positions of two elements to achieve low mutual coupling. It has covered wide frequency bandwidth from 8.47 GHz to 26.76 GHz with higher isolation (more than 22 dB). The overall dimension of this antenna is 35mm × 40mm × 1.6mm. This simulated antenna is realized on FR4 substrate with relative permittivity of 4.4.The ECC (envelope correlation coefficient) and diversity gain are also studied and calculated which provide satisfactory results over the entire operating band for the application of X-band, Ku-band and K-band.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115836849","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}
The deployment of small cells is one of the key enabling technologies for the next generation mobile communication. Due to the short coverage of small cell, the elevation angle, which is ignored in long distance transmission model, should be taken into account in channel modelling. In this paper, we developed a geometrically based single bounce (GBSB) channel model for a small cell mobile environment which assumes that scattering objects are located uniformly within an ellipse around the mobile station (MS). The closed-form expression for joint and marginal probability density function (PDF) of elevation angle of departure (EAOD) is derived and validated by Monte Carlo simulation. With our derived PDF expression, the impact of parameters such as the distance between the base station (BS) and the MS, the height of the BS and the size of scatterers on the elevation angle are also discussed.
{"title":"A Proposed Elevation Angle Model for Small Cell Environments","authors":"Qi Hong, Haonan Hu, Hao Li, Hui Zheng, Baoling Zhang, Jie Zhang","doi":"10.1109/SPIN.2018.8474289","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474289","url":null,"abstract":"The deployment of small cells is one of the key enabling technologies for the next generation mobile communication. Due to the short coverage of small cell, the elevation angle, which is ignored in long distance transmission model, should be taken into account in channel modelling. In this paper, we developed a geometrically based single bounce (GBSB) channel model for a small cell mobile environment which assumes that scattering objects are located uniformly within an ellipse around the mobile station (MS). The closed-form expression for joint and marginal probability density function (PDF) of elevation angle of departure (EAOD) is derived and validated by Monte Carlo simulation. With our derived PDF expression, the impact of parameters such as the distance between the base station (BS) and the MS, the height of the BS and the size of scatterers on the elevation angle are also discussed.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"222 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131516307","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 : 2018-02-01DOI: 10.1109/SPIN.2018.8474187
P. Bai, Kirshna Kumar, Sushil Kumar
Now a days the challenging task in IoT is the communication among each object at every time and everywhere. Limited amount of energy, battery power and processing power of these smart objects lead problems in communication. In the present scenario, energy efficiency is the most prominent requirements that must be served while introducing a protocol for communication in IoT environment. So we need energy efficient communication protocol for Internet of Things at each stage of communication. For this perspective, in this paper we propose energy efficient communication protocol at network layer for Internet of Things. Finally, the performance of the proposed algorithm is compared with respect to the protocol: distributed cluster computing energy efficient routing (DCEER) considering the metric such as routing rounds and no of dead nodes in the variable number of nodes rounds. The final results show that the performance of proposed algorithm is better than the compared algorithm: DCEER in terms of Residual energy and number of rounds.
{"title":"Energy Efficient Communication Protocol at Network Layer for Internet of Things","authors":"P. Bai, Kirshna Kumar, Sushil Kumar","doi":"10.1109/SPIN.2018.8474187","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474187","url":null,"abstract":"Now a days the challenging task in IoT is the communication among each object at every time and everywhere. Limited amount of energy, battery power and processing power of these smart objects lead problems in communication. In the present scenario, energy efficiency is the most prominent requirements that must be served while introducing a protocol for communication in IoT environment. So we need energy efficient communication protocol for Internet of Things at each stage of communication. For this perspective, in this paper we propose energy efficient communication protocol at network layer for Internet of Things. Finally, the performance of the proposed algorithm is compared with respect to the protocol: distributed cluster computing energy efficient routing (DCEER) considering the metric such as routing rounds and no of dead nodes in the variable number of nodes rounds. The final results show that the performance of proposed algorithm is better than the compared algorithm: DCEER in terms of Residual energy and number of rounds.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131770152","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 : 2018-02-01DOI: 10.1109/SPIN.2018.8474055
M. Singhal, Sanyam Shukla
Kernel Extreme Learning Machine (KELM) is used for classification, regression, clustering and feature selection with the help of kernel functions. Conventional KELM uses all training instances as centroids for classification problem while reduced KELM uses randomly choosen training instances as centroids. Furthermore, reduced KELM is used for reducing the computational complexity of conventional KELM. To further improve the computational complexity of KELM, K-means clustering algorithm for centroid selection in KELM is proposed in this paper. In this proposed approach, number of centroids are selected as 1/10 or 5/10 of the total number of training instances and then centroids are computed by using K-means algorithm. Experiments have been carried out by using 15 data sets to illustrate the effectiveness of the proposed method. The results obtained show the reduction in computational time and increment in G-mean which verify the proposed method as an efficient approach in comparison to earlier works.
{"title":"Centroid Selection in Kernel Extreme Learning Machine using K-means","authors":"M. Singhal, Sanyam Shukla","doi":"10.1109/SPIN.2018.8474055","DOIUrl":"https://doi.org/10.1109/SPIN.2018.8474055","url":null,"abstract":"Kernel Extreme Learning Machine (KELM) is used for classification, regression, clustering and feature selection with the help of kernel functions. Conventional KELM uses all training instances as centroids for classification problem while reduced KELM uses randomly choosen training instances as centroids. Furthermore, reduced KELM is used for reducing the computational complexity of conventional KELM. To further improve the computational complexity of KELM, K-means clustering algorithm for centroid selection in KELM is proposed in this paper. In this proposed approach, number of centroids are selected as 1/10 or 5/10 of the total number of training instances and then centroids are computed by using K-means algorithm. Experiments have been carried out by using 15 data sets to illustrate the effectiveness of the proposed method. The results obtained show the reduction in computational time and increment in G-mean which verify the proposed method as an efficient approach in comparison to earlier works.","PeriodicalId":184596,"journal":{"name":"2018 5th International Conference on Signal Processing and Integrated Networks (SPIN)","volume":"53 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132820614","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：