Shing-Tai Pan - Academia.edu (original) (raw)
Papers by Shing-Tai Pan
We propose a recursive singular value decomposition (SVD)-based fuzzy extreme learning machine (R... more We propose a recursive singular value decomposition (SVD)-based fuzzy extreme learning machine (RSVD-F-ELM) for the online learning in classification or regression analysis. By adopting the same architecture and operation as fuzzy extreme learning machine (F-ELM), which is originally designed for the batch learning, and replacing the Moore-Penrose generalized inverse in F-ELM with a recursive SVD-based least squares estimator for optimizing the output weights recursively, RSVD-F-ELM is applicable for the online learning. Compared with the other online learning approach, namely online sequential fuzzy extreme learning machine (OS-F-ELM), experimental results have revealed that RSVD-F-ELM generates the larger accuracy rates in classification analysis and the smaller mean squared errors in regression analysis. Moreover, the learning stability of RSVD-F-ELM is much better.
Recognizing human gender automatically by a computer is a challenging problem. It has been attrac... more Recognizing human gender automatically by a computer is a challenging problem. It has been attracting research attention due to its wide real-life applications. Gender classification can be viewed as an essential preprocessing step in face recognition. Because human faces contain a lot of really useful information, many approaches based on facial features have been investigated for gender classification. In this paper, we present a novel texture pattern as feature descriptor to identify the gender from the facial images. The classification is performed by using a support vector machine. Experimental results on the FERET database are provided to illustrate the proposed approach is an effective method, compared to other similar methods.
Collaborative filtering recommender systems traditionally recommend products to users solely base... more Collaborative filtering recommender systems traditionally recommend products to users solely based on the user-item rating matrix and are simple, convenient to use. In this paper, we focus on two main issues, data sparsity and scalability. Data sparsity can lead to inaccurate recommendations, while scalability may cause an unacceptably long delay before valuable recommendations are acquired. We propose a novel approach to deal with these two issues. Word2Vec is employed to build item vectors from the user comments. Through the user-item rating matrix, user vectors of all the users are then obtained. A clustering technique is applied to reduce the time complexity related to the large numbers of items and users. Experimental results of real data sets are shown to demonstrate the effectiveness of our proposed approach. Keywords—data sparsity, scalability, Word2Vec, selfconstructing clustering, word vectors
Facial expression recognition has received a lot of attention in recent years due to its importan... more Facial expression recognition has received a lot of attention in recent years due to its importance in many multimedia and human-computer interaction applications. One of the critical issues for a successful facial expression recognition system is to develop a discriminative feature descriptor. In this paper, we present a texture descriptor, Local Direction and Transition Pattern, to effectively capture the facial features. The recognition performance of the proposed method is evaluated on the Cohn-Kanade facial expression dataset with a support vector machine classifier. Experimental results show that the proposed method yields good recognition accuracy than other existing methods. KeywordsFacial Expression Recognition, Local Direction and Transition Pattern, Support Vector Machine
Electronics
In recent years, the increasing popularity of smart mobile devices has made the interaction betwe... more In recent years, the increasing popularity of smart mobile devices has made the interaction between devices and users, particularly through voice interaction, more crucial. By enabling smart devices to better understand users’ emotional states through voice data, it becomes possible to provide more personalized services. This paper proposes a novel machine learning model for speech emotion recognition called CLDNN, which combines convolutional neural networks (CNN), long short-term memory neural networks (LSTM), and deep neural networks (DNN). To design a system that closely resembles the human auditory system in recognizing audio signals, this article uses the Mel-frequency cepstral coefficients (MFCCs) of audio data as the input of the machine learning model. First, the MFCCs of the voice signal are extracted as the input of the model. Local feature learning blocks (LFLBs) composed of one-dimensional CNNs are employed to calculate the feature values of the data. As audio signals a...
Full list of author information is available at the end of the article
Advances in Smart Vehicular Technology, Transportation, Communication and Applications, 2018
Human gait is a useful biometric feature for human identification because it can be perceived rem... more Human gait is a useful biometric feature for human identification because it can be perceived remotely without physical contact. One critical step for human gait recognition is to accurately extract visual features. In this paper, we apply the center-symmetric local ternary pattern for feature extraction to identify the person from the gait images. The classification is performed by using a support vector machine. Experiments on the CASIA gait database (Dataset B) are given to illustrate the feasibility of the proposed approach.
Obstacle avoidance is an essential function for the navigation of mobile robots. Noise filtering ... more Obstacle avoidance is an essential function for the navigation of mobile robots. Noise filtering improves the measurement accuracy of senors and plays an important role for obstacle avoidance in the applications of mobile robots. This study evaluates the performance of the extended Kalman filtering (EKF) and Kalman filtering (KF) for obstacle avoidance of a two-wheeled mobile robot. EKF is an advanced version of traditional KF for signal processing. EKF is used to deal with non-linear problems that KF can not process properly and usually has better ability of noise tolerance than KF. Due to the non-linearity and unstability of sensoring results, KF has limited performance in the underlying problem. The robot used in this study carries some sonar sensors that acquire signals of obstacles periodically. EKF linearizes the estimation around the current measure using the partial derivatives of the process and measurement functions to obtain estimates of actual measurements even when non-...
In this paper, the robust stability problem for a class of nominally stable uncertain discrete si... more In this paper, the robust stability problem for a class of nominally stable uncertain discrete singularly perturbed linear systems with multiple time delays is investigated. Stability criteria for the uncertain slow subsystem and the fast subsystem are first derived. A delay-dependent criterion based on spectral norm and spectral radius is then proposed to guarantee the robust stability of the system subject to structured perturbations. 摘 要 在本論文中,我們將探討具多重時延之離散奇 異擾動不確定性系統的強韌穩定性問題。首先,我 們分別提出可使慢速不確定系統與快速不確定系 統漸近穩定之充分條件;其次,我們將另推導出一 充分條件,其可確保原離散奇異擾動不確定系統之
2017 IEEE International Conference on Information and Automation (ICIA), 2017
This study analyzes the effectiveness of the global (the whole face) and local (regions of eyes, ... more This study analyzes the effectiveness of the global (the whole face) and local (regions of eyes, nose, and mouth) features for face recognition. Features describing human faces are encoded in local ternary patterns. The two-class support vector machine is used as the supervised learning algorithm for training recognition models. In the recognition process, recognition modes based on the global features and local features are cascaded. For identifying a face image, the local features are used iteratively for filtering out candidates that can not be clearly identified by the global features, until the one with highest possibility is concluded. The experimental results show that cascading the recognition models of global and local features obtains better classification accuracy than the single classification process.
Fuzzy itemset mining was previously proposed to consider the quantity of items and derive linguis... more Fuzzy itemset mining was previously proposed to consider the quantity of items and derive linguistic rules that are simple and more comprehensible to decision makers. However, most existing fuzzy mining techniques adopt Apriori-based techniques to deal with the problem of mining fuzzy frequent itemsets, and thus their execution efficiency is not good. In this paper, we thus propose an efficient mining approach to speed up the efficiency of finding fuzzy frequent itemsets from databases. In particular, a data-reduction strategy is designed to effectively help prune unpromising fuzzy terms in transactions at each pass in comparison with the other existing algorithms. Through a series of experimental evaluations, the results reveal that the proposed approach runs faster than the existing fuzzy mining algorithms on several synthetic and real datasets under different parameter settings.
IEEE Transactions on Signal Processing, 2009
Electronics Letters, 1996
Advances in Smart Vehicular Technology, Transportation, Communication and Applications, 2018
In this paper, an automatic sleep stages recognition system based on the electrocardiogram (ECG) ... more In this paper, an automatic sleep stages recognition system based on the electrocardiogram (ECG) signals is developed. The reason that ECG signals are used for sleep staging is that the device for measuring ECG signals is cheap and is portable. So, the sleep staging can then be performed at home. In this study, some ECG sleep features used in other research are adopted. These features are used to train the Hidden Markov Model (HMM) model and then fed into the trained HMM for recognition. Unlike the existing research on sleep staging by HMM, in which the modeling of HMM is independent of the special properties of the sleep stage transition, the HMM in this study is adjusted to meet these properties. With this method, the accuracy of sleep staging can be improved. The experimental results show that the proposed method enhances the recognition rate compared with other existing research.
International Journal of Fuzzy Systems, 2016
IEEE Transactions on Instrumentation and Measurement, 2012
ABSTRACT A field-programmable gate array (FPGA)-based robust speech measurement and recognition s... more ABSTRACT A field-programmable gate array (FPGA)-based robust speech measurement and recognition system is the focus of this paper, and the environmental noise problem is its main concern. To accelerate the recognition speed of the FPGA-based speech recognition system, the discrete hidden Markov model is used here to lessen the computation burden inherent in speech recognition. Furthermore, the empirical mode decomposition is used to decompose the measured speech signal contaminated by noise into several intrinsic mode functions (IMFs). The IMFs are then weighted and summed to reconstruct the original clean speech signal. Unlike previous research, in which IMFs were selected by trial and error for specific applications, the weights for each IMF are designed by the genetic algorithm to obtain an optimal solution. The experimental results in this paper reveal that this method achieves a better speech recognition rate for speech subject to various environmental noises. Moreover, this paper also explores the hardware realization of the designed speech measurement and recognition systems on an FPGA-based embedded system with the System-On-a-Chip (SOC) architecture. Since the central-processing-unit core adopted in the SOC has limited computation ability, this paper uses the integer fast Fourier transform (FFT) to replace the floating-point FFT to speed up the computation for capturing speech features through a mel-frequency cepstrum coefficient. The result is a significant reduction in the calculation time without influencing the speech recognition rate. It can be seen from the experiments in this paper that the performance of the implemented hardware is significantly better than that of existing research.
J. Inf. Hiding Multim. Signal Process., 2011
This paper improves the speech recognition speed for the speech recognition chip implemented on a... more This paper improves the speech recognition speed for the speech recognition chip implemented on a FPGA-based embedded system by using Integer Fast Fourier Transform (FFT) on the computing of Mel-Frequency Cepstrum Coefficient (MFCC) for the speech recognition. This paper uses the Hidden Markov Model (HMM) algorithm to construct speech recognition platform. On the embedded system, the computing speed is not as fast as personal computer. This causes that the speech recognition takes much time and power; and further, it does not satisfy the real time requirement. In this paper, we use Integer FFT to replace Float FFT. Experimental results show that the proposed approach reduces much computing time in the speech recognition chip by using Integer FFT with the cost of losing a little recognition rate.
In this paper, an asymptotically stabilizing composite feedback control is proposed for a class o... more In this paper, an asymptotically stabilizing composite feedback control is proposed for a class of linear singularly perturbed systems with multiple time delays. A sufficient condition for the asymptotic stability of the slow subsystem and the fast subsystem is first presented. Moreover, a stability bound ∗ ε of singular perturbation parameter ε is given such that the original system under the composite feedback control is asymptotically stable for all ( ) ∗ ∈ ε ε , 0 . 摘 要 在本論文中,我們將藉由複合回授控制器之設計, 來確保具多重時延之奇異擾動系統之漸近穩定性。首 先,我們分別提出可使慢速與快速子系統為漸近穩定之 充分條件;其次,在慢速與快速子系統為漸近穩定的前 題下,我們另推導出系統之穩定上限值,在此穩定上限 值內,可確保複合回授控制系統之漸近穩定性。
Lecture Notes in Electrical Engineering, 2014
In this paper, we propose a strategy that combines Genetic Algorithm (GA) and HMM to improve the ... more In this paper, we propose a strategy that combines Genetic Algorithm (GA) and HMM to improve the recognition rate of sleep staging. The GA is used to train a better codebook for HMM. With this method, the accuracy and efficiency of sleep medical diagnosis can be expected to be improved. Moreover, some features used in other research are selected as supporting features. These features are used to train the HMM model and then fed into the trained HMM for recognition. Unlike the existing research on sleep staging by HMM, in which the modeling of HMM is independent of the special properties of the sleep stage transition, the HMM in this study is adjusted to meet these properties. The experimental results show that the proposed method greatly enhances the recognition rate compared with other existing research.
We propose a recursive singular value decomposition (SVD)-based fuzzy extreme learning machine (R... more We propose a recursive singular value decomposition (SVD)-based fuzzy extreme learning machine (RSVD-F-ELM) for the online learning in classification or regression analysis. By adopting the same architecture and operation as fuzzy extreme learning machine (F-ELM), which is originally designed for the batch learning, and replacing the Moore-Penrose generalized inverse in F-ELM with a recursive SVD-based least squares estimator for optimizing the output weights recursively, RSVD-F-ELM is applicable for the online learning. Compared with the other online learning approach, namely online sequential fuzzy extreme learning machine (OS-F-ELM), experimental results have revealed that RSVD-F-ELM generates the larger accuracy rates in classification analysis and the smaller mean squared errors in regression analysis. Moreover, the learning stability of RSVD-F-ELM is much better.
Recognizing human gender automatically by a computer is a challenging problem. It has been attrac... more Recognizing human gender automatically by a computer is a challenging problem. It has been attracting research attention due to its wide real-life applications. Gender classification can be viewed as an essential preprocessing step in face recognition. Because human faces contain a lot of really useful information, many approaches based on facial features have been investigated for gender classification. In this paper, we present a novel texture pattern as feature descriptor to identify the gender from the facial images. The classification is performed by using a support vector machine. Experimental results on the FERET database are provided to illustrate the proposed approach is an effective method, compared to other similar methods.
Collaborative filtering recommender systems traditionally recommend products to users solely base... more Collaborative filtering recommender systems traditionally recommend products to users solely based on the user-item rating matrix and are simple, convenient to use. In this paper, we focus on two main issues, data sparsity and scalability. Data sparsity can lead to inaccurate recommendations, while scalability may cause an unacceptably long delay before valuable recommendations are acquired. We propose a novel approach to deal with these two issues. Word2Vec is employed to build item vectors from the user comments. Through the user-item rating matrix, user vectors of all the users are then obtained. A clustering technique is applied to reduce the time complexity related to the large numbers of items and users. Experimental results of real data sets are shown to demonstrate the effectiveness of our proposed approach. Keywords—data sparsity, scalability, Word2Vec, selfconstructing clustering, word vectors
Facial expression recognition has received a lot of attention in recent years due to its importan... more Facial expression recognition has received a lot of attention in recent years due to its importance in many multimedia and human-computer interaction applications. One of the critical issues for a successful facial expression recognition system is to develop a discriminative feature descriptor. In this paper, we present a texture descriptor, Local Direction and Transition Pattern, to effectively capture the facial features. The recognition performance of the proposed method is evaluated on the Cohn-Kanade facial expression dataset with a support vector machine classifier. Experimental results show that the proposed method yields good recognition accuracy than other existing methods. KeywordsFacial Expression Recognition, Local Direction and Transition Pattern, Support Vector Machine
Electronics
In recent years, the increasing popularity of smart mobile devices has made the interaction betwe... more In recent years, the increasing popularity of smart mobile devices has made the interaction between devices and users, particularly through voice interaction, more crucial. By enabling smart devices to better understand users’ emotional states through voice data, it becomes possible to provide more personalized services. This paper proposes a novel machine learning model for speech emotion recognition called CLDNN, which combines convolutional neural networks (CNN), long short-term memory neural networks (LSTM), and deep neural networks (DNN). To design a system that closely resembles the human auditory system in recognizing audio signals, this article uses the Mel-frequency cepstral coefficients (MFCCs) of audio data as the input of the machine learning model. First, the MFCCs of the voice signal are extracted as the input of the model. Local feature learning blocks (LFLBs) composed of one-dimensional CNNs are employed to calculate the feature values of the data. As audio signals a...
Full list of author information is available at the end of the article
Advances in Smart Vehicular Technology, Transportation, Communication and Applications, 2018
Human gait is a useful biometric feature for human identification because it can be perceived rem... more Human gait is a useful biometric feature for human identification because it can be perceived remotely without physical contact. One critical step for human gait recognition is to accurately extract visual features. In this paper, we apply the center-symmetric local ternary pattern for feature extraction to identify the person from the gait images. The classification is performed by using a support vector machine. Experiments on the CASIA gait database (Dataset B) are given to illustrate the feasibility of the proposed approach.
Obstacle avoidance is an essential function for the navigation of mobile robots. Noise filtering ... more Obstacle avoidance is an essential function for the navigation of mobile robots. Noise filtering improves the measurement accuracy of senors and plays an important role for obstacle avoidance in the applications of mobile robots. This study evaluates the performance of the extended Kalman filtering (EKF) and Kalman filtering (KF) for obstacle avoidance of a two-wheeled mobile robot. EKF is an advanced version of traditional KF for signal processing. EKF is used to deal with non-linear problems that KF can not process properly and usually has better ability of noise tolerance than KF. Due to the non-linearity and unstability of sensoring results, KF has limited performance in the underlying problem. The robot used in this study carries some sonar sensors that acquire signals of obstacles periodically. EKF linearizes the estimation around the current measure using the partial derivatives of the process and measurement functions to obtain estimates of actual measurements even when non-...
In this paper, the robust stability problem for a class of nominally stable uncertain discrete si... more In this paper, the robust stability problem for a class of nominally stable uncertain discrete singularly perturbed linear systems with multiple time delays is investigated. Stability criteria for the uncertain slow subsystem and the fast subsystem are first derived. A delay-dependent criterion based on spectral norm and spectral radius is then proposed to guarantee the robust stability of the system subject to structured perturbations. 摘 要 在本論文中,我們將探討具多重時延之離散奇 異擾動不確定性系統的強韌穩定性問題。首先,我 們分別提出可使慢速不確定系統與快速不確定系 統漸近穩定之充分條件;其次,我們將另推導出一 充分條件,其可確保原離散奇異擾動不確定系統之
2017 IEEE International Conference on Information and Automation (ICIA), 2017
This study analyzes the effectiveness of the global (the whole face) and local (regions of eyes, ... more This study analyzes the effectiveness of the global (the whole face) and local (regions of eyes, nose, and mouth) features for face recognition. Features describing human faces are encoded in local ternary patterns. The two-class support vector machine is used as the supervised learning algorithm for training recognition models. In the recognition process, recognition modes based on the global features and local features are cascaded. For identifying a face image, the local features are used iteratively for filtering out candidates that can not be clearly identified by the global features, until the one with highest possibility is concluded. The experimental results show that cascading the recognition models of global and local features obtains better classification accuracy than the single classification process.
Fuzzy itemset mining was previously proposed to consider the quantity of items and derive linguis... more Fuzzy itemset mining was previously proposed to consider the quantity of items and derive linguistic rules that are simple and more comprehensible to decision makers. However, most existing fuzzy mining techniques adopt Apriori-based techniques to deal with the problem of mining fuzzy frequent itemsets, and thus their execution efficiency is not good. In this paper, we thus propose an efficient mining approach to speed up the efficiency of finding fuzzy frequent itemsets from databases. In particular, a data-reduction strategy is designed to effectively help prune unpromising fuzzy terms in transactions at each pass in comparison with the other existing algorithms. Through a series of experimental evaluations, the results reveal that the proposed approach runs faster than the existing fuzzy mining algorithms on several synthetic and real datasets under different parameter settings.
IEEE Transactions on Signal Processing, 2009
Electronics Letters, 1996
Advances in Smart Vehicular Technology, Transportation, Communication and Applications, 2018
In this paper, an automatic sleep stages recognition system based on the electrocardiogram (ECG) ... more In this paper, an automatic sleep stages recognition system based on the electrocardiogram (ECG) signals is developed. The reason that ECG signals are used for sleep staging is that the device for measuring ECG signals is cheap and is portable. So, the sleep staging can then be performed at home. In this study, some ECG sleep features used in other research are adopted. These features are used to train the Hidden Markov Model (HMM) model and then fed into the trained HMM for recognition. Unlike the existing research on sleep staging by HMM, in which the modeling of HMM is independent of the special properties of the sleep stage transition, the HMM in this study is adjusted to meet these properties. With this method, the accuracy of sleep staging can be improved. The experimental results show that the proposed method enhances the recognition rate compared with other existing research.
International Journal of Fuzzy Systems, 2016
IEEE Transactions on Instrumentation and Measurement, 2012
ABSTRACT A field-programmable gate array (FPGA)-based robust speech measurement and recognition s... more ABSTRACT A field-programmable gate array (FPGA)-based robust speech measurement and recognition system is the focus of this paper, and the environmental noise problem is its main concern. To accelerate the recognition speed of the FPGA-based speech recognition system, the discrete hidden Markov model is used here to lessen the computation burden inherent in speech recognition. Furthermore, the empirical mode decomposition is used to decompose the measured speech signal contaminated by noise into several intrinsic mode functions (IMFs). The IMFs are then weighted and summed to reconstruct the original clean speech signal. Unlike previous research, in which IMFs were selected by trial and error for specific applications, the weights for each IMF are designed by the genetic algorithm to obtain an optimal solution. The experimental results in this paper reveal that this method achieves a better speech recognition rate for speech subject to various environmental noises. Moreover, this paper also explores the hardware realization of the designed speech measurement and recognition systems on an FPGA-based embedded system with the System-On-a-Chip (SOC) architecture. Since the central-processing-unit core adopted in the SOC has limited computation ability, this paper uses the integer fast Fourier transform (FFT) to replace the floating-point FFT to speed up the computation for capturing speech features through a mel-frequency cepstrum coefficient. The result is a significant reduction in the calculation time without influencing the speech recognition rate. It can be seen from the experiments in this paper that the performance of the implemented hardware is significantly better than that of existing research.
J. Inf. Hiding Multim. Signal Process., 2011
This paper improves the speech recognition speed for the speech recognition chip implemented on a... more This paper improves the speech recognition speed for the speech recognition chip implemented on a FPGA-based embedded system by using Integer Fast Fourier Transform (FFT) on the computing of Mel-Frequency Cepstrum Coefficient (MFCC) for the speech recognition. This paper uses the Hidden Markov Model (HMM) algorithm to construct speech recognition platform. On the embedded system, the computing speed is not as fast as personal computer. This causes that the speech recognition takes much time and power; and further, it does not satisfy the real time requirement. In this paper, we use Integer FFT to replace Float FFT. Experimental results show that the proposed approach reduces much computing time in the speech recognition chip by using Integer FFT with the cost of losing a little recognition rate.
In this paper, an asymptotically stabilizing composite feedback control is proposed for a class o... more In this paper, an asymptotically stabilizing composite feedback control is proposed for a class of linear singularly perturbed systems with multiple time delays. A sufficient condition for the asymptotic stability of the slow subsystem and the fast subsystem is first presented. Moreover, a stability bound ∗ ε of singular perturbation parameter ε is given such that the original system under the composite feedback control is asymptotically stable for all ( ) ∗ ∈ ε ε , 0 . 摘 要 在本論文中,我們將藉由複合回授控制器之設計, 來確保具多重時延之奇異擾動系統之漸近穩定性。首 先,我們分別提出可使慢速與快速子系統為漸近穩定之 充分條件;其次,在慢速與快速子系統為漸近穩定的前 題下,我們另推導出系統之穩定上限值,在此穩定上限 值內,可確保複合回授控制系統之漸近穩定性。
Lecture Notes in Electrical Engineering, 2014
In this paper, we propose a strategy that combines Genetic Algorithm (GA) and HMM to improve the ... more In this paper, we propose a strategy that combines Genetic Algorithm (GA) and HMM to improve the recognition rate of sleep staging. The GA is used to train a better codebook for HMM. With this method, the accuracy and efficiency of sleep medical diagnosis can be expected to be improved. Moreover, some features used in other research are selected as supporting features. These features are used to train the HMM model and then fed into the trained HMM for recognition. Unlike the existing research on sleep staging by HMM, in which the modeling of HMM is independent of the special properties of the sleep stage transition, the HMM in this study is adjusted to meet these properties. The experimental results show that the proposed method greatly enhances the recognition rate compared with other existing research.