Browsing by Author "Chang, Jyh-Yeong"

Now showing 1 - 20 of 45

A Compensatory NeuroFuzzy System with Online Constructing and Parameter Learning
(2010)
Han, Ming-Feng
;
Lin, Chin-Teng
;
Chang, Jyh-Yeong
;
電控工程研究所
;
Institute of Electrical and Control Engineering
A compensatory neurofuzzy system (CNFS) with on-line learning ability is proposed in this paper. The proposed CNFS model uses a compensatory layer to raise the diversity of fuzzy rules by compensatory weights. The compensatory layer can automatically compare with each fuzzy rule and select higher resources for more important fuzzy rule. An online learning algorithm, which consists of structure learning and parameter learning, is also presented. The structure learning depends on the fuzzy measure to determine the number of fuzzy rules. The parameter learning, based on the gradient descent method, can adjust the shape of the membership function and the weights of the compensatory layer. To demonstrate the capability of the proposed CNFS, it is applied to the Iris, and Wisconsin breast cancer classification datasets from the UCI Repository. Experimental results show that the proposed CNFS for pattern classification can achieve good classification performance.
A Functional-Link based Interval Type-2 Compensatory Fuzzy Neural Network for Nonlinear System modeling
(2011)
Chang, Jyh-Yeong
;
Lin, Yang-Yin
;
Han, Ming-Feng
;
Lin, Chin-Teng
;
電控工程研究所
In this paper, the Functional-Link based Interval Type-2 Compensatory Fuzzy Neural Network (FLIT2CFNN) is a six-layer structure, which combines compensatory fuzzy reasoning method, and the consequent part is combined the proposed functional-link neural network with interval weights. The compensatory fuzzy reasoning method uses adaptive fuzzy operations of neuro-fuzzy systems that can make the fuzzy logic system more adaptive and effective. Initially, there is no rule in the FLIT2CFNN. A FLIT2CFNN is constructed using concurrent structure and parameter learning. The advantages of this learning algorithm are that it converges quickly and the obtained fuzzy rules are more precise. All of the antecedent part parameters and compensatory degree values are learned by gradient descent algorithm. Several simulation results show that the FLIT2CFNN achieves better performance than other feedforword type-1 and type-2 FNNs.
A Fuzzy Weighted Mean Aggregation Algorithm for Color Image Impulse Noise Removal
(2015)
Chang, Jyh-Yeong
;
Liu, Pin-Chang
;
電子工程學系及電子研究所
;
Department of Electronics Engineering and Institute of Electronics
In this paper, we utilize fuzzy weighted mean aggregation algorithm to construct Interval-Valued Fuzzy Relations (IVFR) for grayscale image noise detection. To this end, we use two weighting parameters to calculate the weighted mean difference of the central pixel and its 8-neighborhood pixels in a sliding window across the image. Then, the central pixel will be identified as noisy or non-noisy by using a threshold operation. Besides, to decrease the noise pixel detection error, we have derived an iterative learning mechanism of these weighting parameters of the mean aggregation and thresholds in the training stage. Finally, we embed the pocket algorithm in our learning mechanism to train the best parameter set to minimize the noisy and noise free pixel detection error. The flexibility of the proposed IVFR approach is quite suited to learn the characteristics existing among the noisy pixel and its neighbors. Thus the derived IVPR scheme can excellently detect a noisy pixel and lead to marvelous result on impulsive noise removal. In the pixel restoration stage, we propose a new filtering method. It is divided into three steps: image histogram, noise detection, and image restoration. First, we calculate the histogram of the testing image to find the groups of potential noise pixels. On these possible noisy pixel groups, we make use of the trained weighting parameters to do the fuzzy weighted mean aggregation to double-check whether they are noise corrupted or not. If a pixel is identified as noisy, its value will be restored by a weighted mean filter. Simulation results show that the proposed algorithm provides a significant improvement over other existing filters and preserves more image details. Our algorithm can barely restore the image even when the noise rate is as high as 97 %.
A Hybrid of Cuckoo Search and Simplex Method for Fuzzy Neural Network Training
(2015)
Chang, Jyh-Yeong
;
Liao, Shih-Hui
;
Wu, Shang-Lin
;
Lin, Chin-Teng
;
電控工程研究所
In this paper, a new hybrid algorithm mixing the simplex method of Nelder and Mead (NM) and the cuckoo search (CS), abbreviated as NM-CS, is proposed for the training of the Fuzzy Neural Networks (FNNs). In standard CS, cuckoo birds engage the obligate brood parasitism by laying their own eggs to other host birds. If a host bird discovers the alien eggs, they will either throw these eggs away or abandon its nest and build a new nest elsewhere. In the proposed hybrid algorithm, instead of using the probability to discover an alien egg for the CS, we use the concept of a simplex which is used in the NM algorithm to abandon and generate the new nests. Our proposed method puts more emphasis on exploration of the search space and enhances the ability to avoid local optimum. Some simulation problems will be provided to compare the performances of the proposed method and other methods in training an FNN. In these simulations, it is observed that the proposed method outperforms other methods.
A Mutually Recurrent Interval Type-2 Neural Fuzzy System (MRIT2NFS) With Self-Evolving Structure and Parameters
(2013-06-01)
Lin, Yang-Yin
;
Chang, Jyh-Yeong
;
Pal, Nikhil R.
;
Lin, Chin-Teng
;
電控工程研究所
In this paper, a mutually recurrent interval type-2 neural fuzzy system (MRIT2NFS) is proposed for the identification of nonlinear and time-varying systems. The MRIT2NFS uses type-2 fuzzy sets in order to enhance noise tolerance of the system. In the MRIT2NFS, the antecedent part of each recurrent fuzzy rule is defined using interval type-2 fuzzy sets, and the consequent part is of the Takagi-Sugeno-Kang type with interval weights. The antecedent part of MRIT2NFS forms a local internal feedback and interaction loop by feeding the rule firing strength of each rule to others including itself. The consequent is a linear combination of exogenous input variables. The learning of MRIT2NFS starts with an empty rule base and all rules are learned online via structure and parameter learning. The structure learning of MRIT2NFS uses online type-2 fuzzy clustering. For parameter learning, the consequent part parameters are tuned by rule-ordered Kalman filter algorithm to reinforce parameter learning ability. The type-2 fuzzy sets in the antecedent and weights representing the mutual feedback are learned by the gradient descent algorithm. After the training, a weight-elimination scheme eliminates feedback connections that do not have much effect on the network behavior. This method can efficiently remove redundant recurrence and interaction weights. Finally, the MRIT2NFS is used for system identification under both noise-free and noisy environments. For this, we consider both time series prediction and nonlinear plant modeling. Compared with type-1 recurrent fuzzy neural networks, simulation results show that our approach produces smaller root-mean-squared errors using the same number of iterations.
A new two-DOF rotational optical image stabilizer
(2011-06-01)
Chao, Paul C-P
;
Chen, Yu-Han
;
Chiu, Chi-Wei
;
Tsai, Meng-Yen
;
Chang, Jyh-Yeong
This work proposes design, optimization and validation of a new two degree-of-freedom (DOF) rotational optical image stabilizer (OIS) that is suitable for installing into a mobile-phone-camera. This OIS differs from the previous designs since it stabilizes the lens holder directly by two-DOF translational mechanisms that are actuated by voice coil motors. The work can be divided into three main parts: (1) designs of mechanism and genetic algorithm (GA) optimization of magnetic field, (2) establishments of the dynamic model and equation of motions (EOM) of a dual-axis rotational structure, and (3) realizations of a sliding mode control (SMC) controller with fine performance. The dynamic of the dual-axis rotational OIS system has been analyzed and the EOM has been derived. Based on the Lagrange's method, the motions of the OIS have been modeled through considering kinetic energy and electromagnetic torques. In the last part, the theory of SMC is applied, and the associated simulations are conducted. Based on the simulation results, the SMC controller is forged with the assistances from MATLAB pre-simulation and tested by a microprocessor module. After a series of experiments and verifications, the prototype of the novel OIS is finally accomplished with satisfactory performance of vibration reduction.
A Novel Two-Stage Impulse Noise Removal Technique Based on Neural Networks and Fuzzy Decision
(2008-08-01)
Liang, Sheng-Fu
;
Lu, Shih-Mao
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng (CT)
;
交大名義發表
In this paper, a novel two-stage noise removal algorithm to deal with impulse noise is proposed. In the first stage, an adaptive two-level feedforward neural network (NN) with a back-propagation training algorithm was applied to remove the noise cleanly and keep the uncorrupted information well. In the second stage, the fuzzy decision rules inspired by the human visual system (HVS) are proposed to classify the image pixels into human perception sensitive class and nonsensitive class, and to compensate the blur of the edge and the destruction caused by the median filter. An NN is proposed to enhance the sensitive regions with higher visual quality. According to the experimental results, the proposed method is superior to conventional methods in perceptual image quality as well as the clarity and smoothness in edge regions.
A TSK-Type-Based Self-Evolving Compensatory Interval Type-2 Fuzzy Neural Network (TSCIT2FNN) and Its Applications
(2014-01-01)
Lin, Yang-Yin
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng
;
電控工程研究所
;
腦科學研究中心
In this paper, a Takagi-Sugeno-Kang (TSK)-type-based self-evolving compensatory interval type-2 fuzzy neural network (FNN) (TSCIT2FNN) is proposed for system modeling and noise cancellation problems. A TSCIT2FNN uses type-2 fuzzy sets in an FNN in order to handle the uncertainties associated with information or data in the knowledge base. The antecedent part of each compensatory fuzzy rule is an interval type-2 fuzzy set in the TSCIT2FNN, where compensatory-based fuzzy reasoning uses adaptive fuzzy operation of a neural fuzzy system to make the fuzzy logic system effective and adaptive, and the consequent part is of the TSK type. The TSK-type consequent part is a linear combination of exogenous input variables. Initially, the rule base in the TSCIT2FNN is empty. All rules are derived according to online type-2 fuzzy clustering. For parameter learning, the consequent part parameters are tuned by a variable-expansive Kalman filter algorithm to the reinforce parameter learning ability. The antecedent type-2 fuzzy sets and compensatory weights are learned by a gradient descent algorithm to improve the learning performance. The performance of TSCIT2FNN for the identification is validated and compared with several type-1 and type-2 FNNs. Simulation results show that our approach produces smaller root-mean-square errors and converges more quickly.
Adaptive Least Trimmed Squares Fuzzy Neural Network
(2012)
Chang, Jyh-Yeong
;
Liao, Shih-Hui
;
Lin, Chin-Teng
;
電控工程研究所
;
Institute of Electrical and Control Engineering
In this paper, we propose the adaptive least trimmed squares fuzzy neural network (ALTS-FNN), which applies the scale estimate to the least trimmed squares fuzzy neural network (LTS-FNN). The emphasis of this paper is particular on the robustness against the outliers and the choice of the trimming constant can be determined adaptively. Some numerical examples will be provided to compare the robustness against outliers for usual FNN and the ALTS-FNN. Simulation results show that the ALTS-FNN in the paper have good performance for outlier detection.
Adaptive neural-based fuzzy modeling for biological systems
(2013-04-01)
Wu, Shinq-Jen
;
Wu, Cheng-Tao
;
Chang, Jyh-Yeong
;
電控工程研究所
;
Institute of Electrical and Control Engineering
The inverse problem of identifying dynamic biological networks from their time-course response data set is a cornerstone of systems biology. Hill and Michaelis-Menten model, which is a forward approach, provides local kinetic information. However, repeated modifications and a large amount of experimental data are necessary for the parameter identification. S-system model, which is composed of highly nonlinear differential equations, provides the direct identification of an interactive network. However, the identification of skeletal-network structure is challenging. Moreover, biological systems are always subject to uncertainty and noise. Are there suitable candidates with the potential to deal with noise-contaminated data sets? Fuzzy set theory is developed for handing uncertainty, imprecision and complexity in the real world; for example, we say "driving speed is high" wherein speed is a fuzzy variable and high is a fuzzy set, which uses the membership function to indicate the degree of a element belonging to the set (words in Italics to denote fuzzy variables or fuzzy sets). Neural network possesses good robustness and learning capability. In this study we hybrid these two together into a neural-fuzzy modeling technique. A biological system is formulated to a multi-input-multi-output (MIMO) Takagi-Sugeno (T-S) fuzzy system, which is composed of rule-based linear subsystems. Two kinds of smooth membership functions (MFs), Gaussian and Bell-shaped MFs, are used. The performance of the proposed method is tested with three biological systems. (C) 2013 Elsevier Inc. All rights reserved.
An Internal/Interconnection Recurrent Type-2 Fuzzy Neural Network (IRT2FNN) for Dynamic System Identification
(2010)
Lin, Yang-Yin
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng
;
電控工程研究所
;
Institute of Electrical and Control Engineering
This paper proposes an Internal/Interconnection Recurrent Type-2 Fuzzy Neural Network (IRT2FNN) for dynamic system identification. The antecedent part of IRT2FNN forms a self and interconnection feedback loop by feeding the past and current firing strength of each rule. The TSK-type consequent part is a linear model of exogenous inputs with interval weights. The initial rule base in the IRT2FNN is empty, and an on-line constructing method is proposed to generate fuzzy rules which flexibly partition the input space. The recurrent structure in the IRT2FNN enable to handle dynamic system identification problems with a priori knowledge of system input and output delay numbers. Simulations on dynamic system identification verify the performance of IRT2FNN with clean and noisy outputs as well.
An Interval Type-2 Neural Fuzzy Inference System (IT2NFIS) with compensatory operator
(2013)
Lin, Yang-Yin
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng
;
電控工程研究所
;
Institute of Electrical and Control Engineering
In this paper, an interval type-2 neural fuzzy system (IT2NFIS) with compensatory operator is proposed for system modeling. The IT2NFIS uses type-2 fuzzy sets in the premise clause in order to effectively handle the uncertainties in terms of data and information. The premise part of each compensatory fuzzy rule is an interval type-2 fuzzy set in the IT2NFIS, where compensatory operation is able to adaptively adjust fuzzy membership functions and to dynamically optimize fuzzy operations. The consequent part in the IT2NFIS consists of the Takagi-Sugeno-Kang (TSK) type that is a linear combination of exogenous input variables. Initially the rule base in the IT2NFIS is empty. All rules generated are based on on-line type-2 fuzzy clustering. All free weights are learned by a gradient descent algorithm to improve the learning performance. Simulation results show that our approach yields smaller root mean squared errors than its rivals.
APPLYING GENERALIZED WEIGHTED MEAN AGGREGATION TO IMPULSIVE NOISE REMOVAL OF IMAGES
(2014)
Chen, Kuan-Lin
;
Chang, Jyh-Yeong
;
電機工程學系
;
Department of Electrical and Computer Engineering
In this paper, we apply generalized weighted mean to construct interval-valued fuzzy relations for grayscale image impulse noise detection and correction. First, we employ two weighting parameters and perform the weighted mean aggregation for the central pixel and its eight neighbor pixels in a 3x3 sliding window across the image. Then, to counter the over-weighting of a big difference term, we apply a saturation threshold transfer function to these eight pixel difference values. Finally, the image noise map is obtained through a threshold operation on the cumulative differences. To decrease the noise detection error, weighting parameters of the mean can be learned by the gradient method caste in discrete formulation. Moreover, to get higher PSNR in the corrected image, we have experienced from the training that we will select weight of 20 for noise rate smaller than 20% and 50 for noise rate greater than 20%, on erroneous noisy than that on the erroneous non-noise pixel. By the experiment, we have shown that the integration of interval-valued fuzzy relations with the weighted mean aggregation algorithm can effectively detect the image noise pixels and then correct them thereafter.
Applying Weighted Mean Aggregation to Edge Detection of Images
(2013)
Chang, Jyh-Yeong
;
Chang, Yi-Hsin
;
電子工程學系及電子研究所
;
Department of Electronics Engineering and Institute of Electronics
This paper applies weighted mean to construct interval-valued fuzzy relations for grayscale image edge detection. This fuzzy relation image shows the changes in intensity values between a 3x3 window central pixel and its eight neighbor pixels. We employ two weighting parameters, and perform the weighted mean aggregation for the central pixel and its eight neighbor pixels in a sliding window across the image to lead to the fuzzy edge images. Finally, the image edge map is obtained through a threshold operation. Moreover, to decrease the edge detection error, weighting parameters of the mean can be learned by the gradient method caste in discrete formulation. By the training results of eight grayscale synthetic images with adding random noises, we have shown that the integration of interval-valued fuzzy relations with the weighted mean aggregation algorithm will produce a more robust response in detecting the image edge. Finally, by applying the optimal edge detection parameters to natural images, we have found that it is better compared to the well-known Canny edge detector.
Automatic Age Estimation System for Face Images Regular Paper
(2012-11-28)
Lin, Chin-Teng
;
Li, Dong-Lin
;
Lai, Jian-Hao
;
Han, Ming-Feng
;
Chang, Jyh-Yeong
Humans are the most important tracking objects in surveillance systems. However, human tracking is not enough to provide the required information for personalized recognition. In this paper, we present a novel and reliable framework for automatic age estimation based on computer vision. It exploits global face features based on the combination of Gabor wavelets and orthogonal locality preserving projections. In addition, the proposed system can extract face aging features automatically in real-time. This means that the proposed system has more potential in applications compared to other semi-automatic systems. The results obtained from this novel approach could provide clearer insight for operators in the field of age estimation to develop real-world applications.
Biosensor Technologies for Augmented Brain-Computer Interfaces in the Next Decades
(2012-05-01)
Liao, Lun-De
;
Lin, Chin-Teng
;
McDowell, Kaleb
;
Wickenden, Alma E.
;
Gramann, Klaus
The study of brain-computer interfaces (BCIs) has undergone 30 years of intense development and has grown into a rich and diverse field. BCIs are technologies that enable direct communication between the human brain and external devices. Conventionally, wet electrodes have been employed to obtain unprecedented sensitivity to high-temporal-resolution brain activity; recently, the growing availability of various sensors that can be used to detect high-quality brain signals in a wide range of clinical and everyday environments is being exploited. This development of biosensing neurotechnologies and the desire to implement them in real-world applications have led to the opportunity to develop augmented BCIs (ABCIs) in the upcoming decades. An ABCI is similar to a BCI in that it relies on biosensors that record signals from the brain in everyday environments; the signals are then processed in real time to monitor the behavior of the human. To use an ABCI as a mobile brain imaging technique for everyday, real-life applications, the sensors and the corresponding device must be lightweight and the equipment response time must be short. This study presents an overview of the wide range of biosensor approaches currently being applied to ABCIs, from their use in the laboratory to their application in clinical and everyday use. The basic principles of each technique are described along with examples of current applications of cutting-edge neuroscience research. In summary, we show that ABCI techniques continue to grow and evolve, incorporating new technologies and advances to address ever more complex and important neuroscience issues, with advancements that are envisioned to lead to a wide range of real-life applications.
Day-and-Night Video Based Face Identification
(2015)
Chang, Jyh-Yeong
;
Chan, Tzu-Rou
;
Fu, Hsin-Chia
;
電子工程學系及電子研究所
;
Department of Electronics Engineering and Institute of Electronics
Human face recognition system is a desired technique in our daily life. It is a widely well-come technique that can all-day-long and on-line recognize a person from video cameras. To this end, we use a near infrared (NIR) camera to capture day-and-night video images for on-line human recognition. In this paper, we adopt human face sub-image attraction package in OpenCV, which is based on Haar cascade classifier. The package is a feature-based algorithm and works much faster than the pixel-based algorithm. It is to be noted that the image contrast color tones of video frames in the night is worse than that in the day, thus we employ multi-scale retinex to enhance video frames in the night before OpenCV face extraction routine. The extracted face sub-image is first transformed to a new space by eigenspace and canonical space transformation. The recognition is finally done in canonical space. Despite OpenCV\'s popularity to date, extracting face sub-images from taken videos are still not reliable enough. Namely, we can obtain many non-face sub-images among the extracted face sub-images. We judiciously classify the sub-images that are far away from the centroids of persons to be classified as non-face sub-images. This may remedy the shortcoming of OpenCV package, and greatly increase the face recognition accuracy. Furthermore, we consider the most recent three consecutive face image recognitions from video, and use majority vote to recognize a person to enhance the accuracy. Besides, we have tested face image recognition to reject intruders successfully.
Design, Fabrication and Experimental Validation of a Novel Dry-Contact Sensor for Measuring Electroencephalography Signals without Skin Preparation
(2011-06-01)
Liao, Lun-De
;
Wang, I-Jan
;
Chen, Sheng-Fu
;
Chang, Jyh-Yeong
;
Lin, Chin-Teng
In the present study, novel dry-contact sensors for measuring electro-encephalography (EEG) signals without any skin preparation are designed, fabricated by an injection molding manufacturing process and experimentally validated. Conventional wet electrodes are commonly used to measure EEG signals; they provide excellent EEG signals subject to proper skin preparation and conductive gel application. However, a series of skin preparation procedures for applying the wet electrodes is always required and usually creates trouble for users. To overcome these drawbacks, novel dry-contact EEG sensors were proposed for potential operation in the presence or absence of hair and without any skin preparation or conductive gel usage. The dry EEG sensors were designed to contact the scalp surface with 17 spring contact probes. Each probe was designed to include a probe head, plunger, spring, and barrel. The 17 probes were inserted into a flexible substrate using a one-time forming process via an established injection molding procedure. With these 17 spring contact probes, the flexible substrate allows for high geometric conformity between the sensor and the irregular scalp surface to maintain low skin-sensor interface impedance. Additionally, the flexible substrate also initiates a sensor buffer effect, eliminating pain when force is applied. The proposed dry EEG sensor was reliable in measuring EEG signals without any skin preparation or conductive gel usage, as compared with the conventional wet electrodes.
Design, simulation and experimental validation of a novel flexible neural probe for deep brain stimulation and multichannel recording
(2012-06-01)
Lai, Hsin-Yi
;
Liao, Lun-De
;
Lin, Chin-Teng
;
Hsu, Jui-Hsiang
;
He, Xin
An implantable micromachined neural probe with multichannel electrode arrays for both neural signal recording and electrical stimulation was designed, simulated and experimentally validated for deep brain stimulation (DBS) applications. The developed probe has a rough three-dimensional microstructure on the electrode surface to maximize the electrode-tissue contact area. The flexible, polyimide-based microelectrode arrays were each composed of a long shaft (14.9 mm in length) and 16 electrodes (5 mu m thick and with a diameter of 16 mu m). The ability of these arrays to record and stimulate specific areas in a rat brain was evaluated. Moreover, we have developed a finite element model (FEM) applied to an electric field to evaluate the volume of tissue activated (VTA) by DBS as a function of the stimulation parameters. The signal-to-noise ratio ranged from 4.4 to 5 over a 50 day recording period, indicating that the laboratory-designed neural probe is reliable and may be used successfully for long-term recordings. The somatosensory evoked potential (SSEP) obtained by thalamic stimulations and in vivo electrode-electrolyte interface impedance measurements was stable for 50 days and demonstrated that the neural probe is feasible for long-term stimulation. A strongly linear (positive correlation) relationship was observed among the simulated VTA, the absolute value of the SSEP during the 200 ms post-stimulus period (Sigma SSEP) and c-Fos expression, indicating that the simulated VTA has perfect sensitivity to predict the evoked responses (c-Fos expression). This laboratory-designed neural probe and its FEM simulation represent a simple, functionally effective technique for studying DBS and neural recordings in animal models.
Designing Mamdani Type Fuzzy Rule Using a Collaborative FCM Scheme
(2013)
Lin, Chin-Teng
;
Prasad, Mukesh
;
Chang, Jyh-Yeong
;
電子工程學系及電子研究所
;
Department of Electronics Engineering and Institute of Electronics
This paper presents a new approach for generating fuzzy rules for fuzzy inference system by using collaborative fuzzy c-mean (CFCM). In order to do any mode of integration between datasets, there is a need to define the common feature between datasets by using some kind of collaborative process and also need to preserve the privacy and security at higher levels. This collaboration process gives a common structure between datasets which helps to define an appropriate number of rules for structural learning and also improve the accuracy of the system modeling. This all consideration bring the concept of collaborative fuzzy rule generation process with a quality measuring.