Övünç Polat

Hand geometry identification without feature extraction by general regression neural network

This paper presents an approach to automatically recognize hand geometry pattern based on a database. The system does not require any feature extraction stage before the identification. General regression neural networks are used for the classification and/or verification of the patterns. Simulation results show that hand geometry pattern identification by the proposed method improves the identification rate considerably. To show the system performance, false acceptance ratio and false rejection ratio are given.

Genetic optimization of GRNN for pattern recognition without feature extraction

This paper describes an approach for pattern recognition using genetic algorithm and general regression neural network (GRNN). The designed system can be used for both 3D object recognition from 2D poses of the object and handwritten digit recognition applications. The system does not require any preprocessing and feature extraction stage before the recognition. In GRNN, placement of centers has significant effect on the performance of the network. The centers and widths of the hidden layer neuron basis functions are coded in a chromosome and these two critical parameters are determined by the optimization using genetic algorithms. Experimental results show that the optimized GRNN provides higher recognition ability compared with that of unoptimized GRNN.

FPGA implementation of a General Regression Neural Network: An embedded pattern classification system

This study proposes an approach to implement a General Regression Neural Network (GRNN) based on Field Programmable Gate Array (FPGA). The GRNN has a four-layer structure which is comprised of an input layer, a pattern layer, a summation layer and an output layer. The layers of GRNN are designed with fixed-point arithmetic using synthesizable VHDL (Very High Speed Integrated Circuit Hardware Description Language) code for FPGA implementation. In this work, the system was designed for pattern classification applications; however, it can be used for other application areas of GRNN. Different datasets were used to test the GRNN. Simulation results show that pattern classification by hardware implementation of GRNN has successfully achieved. The proposed system is flexible and scalable. For different classification applications, it can be modified easily according to number of inputs and number of reference data.

Determination of sympathovagal balance in ventricular tachiarrythmia patients with implanted cardioverter defibrillators using wavelet transform and MLPNN

HRV is a nonstationary signal that includes sympathovagal balance (SB) information related to LF/HF ratio between the sympathetic and parasympathetic nervous systems. In this paper, a solution based on Daubechies wavelet transform (dbN) and multilayer perceptron neural network (MLPNN) has been presented for the determination of SB. HRV database obtained MIT-BIH arrhythmia database consisting of pairs of RR interval time series, recorded by implanted cardioverter defibrillators in 78 subjects. RMS values of approximation and detail components (Arms and Drms) obtained from dbN wavelet transform of HRV signals have been used as training data for MLPNN. Trains were realized in 5 different dbN with only Arms components, only Drms components and both of them and results were compared. Train accuracy and test accuracy results have been reached very successful percentage values that might be valuable for clinical applications.

Determination of a New VLF Band in HRV for Ventricular Tachyarrhythmia Patients

This study presents a new very low frequency (VLF) band range in ventricular tachyarrhythmia patients and involves an approach for estimation of effect of VLF band on ventricular tachyarrhythmia patients. A model based on wavelet packets (WP) and multilayer perceptron neural network (MLPNN) is used for determination of effective VLF band in heart rate variability (HRV) signals. HRV is decomposed into sub-bands including very low frequency parts and variations of energy are analyzed. Domination test is done using MLPNN and dominant band is determined. As a result, a new VLF band was described in 0.0039063–0.03125 Hz frequency range. This method can be used for other bands or other arrhythmia patients. Especially, estimation of dominant band energy using this method can be helped to diagnose for applications where have important effect of characteristic band.

Genetic optimizations for radial basis function and general regression neural networks

The topology of a neural network has a significant importance on the networks performance. Although this is well known, finding optimal configurations is still an open problem. This paper proposes a solution to this problem for Radial Basis Function (RBF) networks and General Regression Neural Network (GRNN) which is a kind of radial basis networks. In such networks, placement of centers has significant effect on the performance of network. The centers and widths of the hidden layer neuron basis functions are coded in a chromosome and these two critical parameters are determined by the optimization using genetic algorithms. Thyroid, iris and escherichia coli bacteria datasets are used to test the algorithm proposed in this study. The most important advantage of this algorithm is getting succesful results by using only a small part of a benchmark. Some numerical solution results indicate the applicability of the proposed approach.

Estimation and evaluation of sub-bands on LF and HF base-bands in HRV for Ventricular Tachyarrhythmia patients

Heart Rate Variability (HRV) is an efficient tool for assessment of Sympathovagal Balance (SB) and classification of cardiac disturbances. However, its index may be not enough for classification and evaluation of some disease. This study presents 32 new sub-bands over LF and HF base-bands that are accepted in the literature. Moreover, it determines dominant sub-bands over both base-bands in VTA database. These sub-bands are obtained using Wavelet Packet Transform (WPT) and evaluated using Multilayer Perceptron Neural Networks (MLPNN). Results are compared with obtained results from normal datasets. The domination effects of these sub-bands are assessed according to comparison of each other related to MLPNN training and test accuracy percentages by selecting different width of windows. As a result, obtained results showed that the LF zone including LF1, LF2 and LF3 sub-bands on 0.0390625-0.0859375Hz frequency range is the most dominant over the LF base-band and, the HF zone including HF1, HF2 and HF3 on 0.1953125-0.28125Hz frequency range is the most dominant over the HF base-band. In normal datasets, distinctive domination effect has not been determined.

Recognition of Patterns Without Feature Extraction by GRNN

Automatic pattern recognition is a very important task in many applications such as image segmentation, object detection, etc. This work aims to find a new approach to automatically recognize patterns such as 3D objects and handwritten digits based on a database using General Regression Neural Networks (GRNN). The designed system can be used for both 3D object recognition from 2D poses of the object and handwritten digit recognition applications. The system does not require any preprocessing and feature extraction stage before the recognition. Simulation results show that pattern recognition by GRNN improves the recognition rate considerably in comparison to other neural network structures and has shown better recognition rates and much faster training times than that of Radial Basis Function and Multilayer Perceptron networks for the same applications.

3-D object recognition using 2-d poses processed by CNNs and a GRNN

This paper presents a novel approach to automatically recognize objects. The system used is a new model that contains two blocks; one for extracting direction and pixel features from object images using Cellular Neural Networks (CNN), and the other for classification of objects using a General Regression Neural Network (GRNN). A data set consisting of different properties of 10 different objects is prepared by CNN.

Design and analysis of low power Carbon Nanotube Field Effect Transistor (CNFET) D Flip-Flops (DFFs)

This paper presents design and analysis of D Flip-Flops (DFFs) using Carbon Nanotube Field-Effect Transistors (CNFETs). Two different DFF circuits are implemented. Circuit performance of CNFET models have been compared to silicon based CMOS models in terms of Clk-Q delay, average power, power delay product (PDP), setup time, hold time, minimum operating voltage, area and average leakage power. CNFET DFFs have shown superior performance over CMOS DFFs in simulations for all the performance parameters.