Majlesi Journal of Electrical Engineering
Year:2010 | Volume:4 | Issue:4 (15)|Papers Count:8

Emotion has an important role in naturalness of man-machine communication and many researchers investigate computerized emotion recognition from speech in the recent decades. In this paper, the effect of formant-related features on improving the performance of emotion detection systems is experimented. To do this, various forms and combinations of the first three formants are concatenated to a popular feature vector and Gaussian mixture models are used as classifiers. Experimental results show average recognition rate of 69% in four emotional states and noticeable performance improvement by adding only one formant-related parameter to feature vector. The architecture of hybrid emotion recognition/spotting is also proposed based on the developed models.

Jaccard index similarity measure, which applies the extension principle approach to obtain fuzzy maximum and fuzzy minimum, has been proposed in ranking fuzzy numbers. However, the extension principle used is only applicable to normal fuzzy numbers and, therefore, fails to rank non-normal ones. Apart from that, the extension principle does not preserve the type of membership function of fuzzy numbers and also involves laborious mathematical operations. In this paper, a simple vertex fuzzy arithmetic operation, namely the function principle is applied. This paper also proposes the degree of optimism concept in aggregating the fuzzy evidence. The method is capable to rank both normal and non-normal fuzzy numbers in a simpler manner from all decision makers’ perspectives.

The ever-increasing number of logo (trademark) in official automation systems for information management, archiving and retrieval applications has created greater demand for an automatic detection and recognition logo. In this paper, a hierarchical classification structure based on Bayesian classifier is proposed to logo detection and recognition. In this hierarchical structure, using two measures false accept rate (FAR) and false reject rate (FRR), a novel and straightforward training scheme is presented to extract optimum parameters of the trained Bayesian classifier. In each level of the hierarchical structure, a separable feature set of shape and texture features is used to train and test classifier based on complexity of the logo pattern. The logo candidate regions are extracted from document images by a wavelet-based segmentation algorithm, and then recognized in the proposed structure. The proposed structure is evaluated on a vast database consisting of the document and non-document images with Persian and international logos. The obtained results show efficiency of the proposed structure in the real and operational conditions.

Knowledge embedded within artificial neural networks (ANNs) is distributed over the connections and weights of neurons. So, the user considers ANN as a black box system. There are many researches investigating the area of rule extraction by ANNs. In this paper, a dynamic cell structure (DCS) neural network and a modified version of LERX algorithm are used for rule extraction. On the other hand, intrusion detection system (IDS) is known as a critical technology to secure computer networks. So, the proposed algorithm is used to develop IDS and classify the patterns of intrusion. To compare the performance of the proposed system with other machine learning algorithms, multi-layer perceptron (MLP) with output weight optimization-hidden weight optimization (OWO-HWO) training algorithm is employed with selected inputs based on the results of a feature relevance analysis. Empirical results show the superior performance of the IDS based on rule extraction from DCS, in recognizing hard-detectable attack categories, e.g. userto- root (U2R) and also offering competitive false alarm rate (FAR). Although, MLP with 25 selected input features, instead of 41 standard features introduced by knowledge discovery and data mining group (KDD), performs better in terms of detection rate (DR) and cost per example (CPE) when compared with some other machine learning methods, as well.

The direct force thrust control (DTFC) is linear type of the direct torque control (DTC) method. The advantages of DTFC method are structure simplicity, low dependency to motor parameters and no requirement to coordination transformations. In this paper this method is modified in order to eliminate the defects that include the switching frequency and exciting large ripples of force and flux. In previous works, the structure simplicity of DTC, rare calculations to reduce the force ripples and fixing switching frequency are disaffirmed. With regards to keeping DTC advantages, a new method is presented in this paper to eliminate the defects by the aid of neural network. Also, the precise non-linear behavior of PMLSM motor in DTC has been considered by using space vector modulation. Finally, the simulation results concluded by the submitted intelligent DTC-SVM method are more satisfactory than other methods.

In a permanent magnet (PM) linear motor, there is a force ripple which is detrimental to positioning. This force ripple is mainly due to a cogging force and a mutual force ripple. These forces are affected by the geometric parameters of a brushless PM motor, such as the width of the magnet, the height of the magnet, the shifted ength of the magnetic pole, the length and height of the armature and the slot width. The optimal design can be found by considering force ripple as a cost function and the geometric parameters as design variables. In this paper, we calculate the flux density distribution in the air gap using the analytic solution of Laplace and Possion equations in the function of geometric parameters. The cogging force is obtained by integrating the Maxwell stress tensor, which is described by the flux density distribution on the slot face and end face of the iron core of an armature. Finally, a finite element method is presented in order to compare with the previous method.

This paper presents the comparative performance of neuro- Fuzzy controlled Voltage Source Converters (VSC) based Flexible AC Transmission System (FACTS) devices, such as Static Synchronous Series Compensator (SSSC), Static Synchronous Compensator (STATCOM), and Unified Power Flow Controller (UPFC) in terms of improvement in transient stability. In neuro-fuzzy control method the simplicity of fuzzy systems and the ability of training in neural networks have been combined. The training data set the parameters of membership functions in fuzzy controller. This Adaptive Network Fuzzy Inference System (ANFIS) can track the given input-output data in order to conform to the desired controller. The maximization of energy function of UPFC is used as an objective function to generate the training data. Proposed method is tested on a single machine infinitive bus system to confirm its performance through simulation. The results prove the noticeable influence of ANFIS controlled UPFC on increasing Critical Clearing Time (CCT) of system.

This paper presents the study of vocal videostroboscopic videos to detect morphological pathologies using an active contour segmentation and objective measurements. The ad-hoc designed active contour algorithm permits to obtain a robust and fast segmentation using vocal folds images in RGB format. In this work, we have employed connected component analysis as a post-processing tool. After the segmentation process the image is analyzed and the pathology can be localized automatically and we can extract some features of each fold (such as the size of the polyp or cyst, the glottal space, the position…). Experimental results demonstrate that the proposed method is effective. Our proposal segments correctly the 95% of database test videos and it shows a great advance in design. The objective measurements complete a new method to diagnose vocal folds pathologies.