Radek Holota | University of West Bohemia (original) (raw)
Papers by Radek Holota
Journal of Real-Time Image Processing, 2013
This paper deals with the design and the implementation of an image recognition system based on F... more This paper deals with the design and the implementation of an image recognition system based on FPGA devices. It explores an n-tuple methodology using node ‘grouping’ and the possible advantages offered by this little-known technique. The paper is based on the implementation of this concept by an FPGA device. A novel approach to the organization of the neural networks data in the n-tuple memory is introduced. The system was tested on a real-world recognition task—the recognition of road signs. The test results are presented and discussed. It is concluded that the designed system may be a powerful part of more complex equipment for the solution of many recognition issues.
This article deals with a design of CAN bus analyser. For this purpose the design with utilisatio... more This article deals with a design of CAN bus analyser. For this purpose the design with utilisation of Field Programmable Gate Arrays (FPGAs) is used. A system structure consisted of basic blocks is introduced and the blocks are described. For the structure description is used VHDL (VHSIC Hardware Description Language).
Oriented structures in metallography are usually evaluated by the intersection method or on the b... more Oriented structures in metallography are usually evaluated by the intersection method or on the base of comparison against a series of standard micrographs. In this paper is introduced a new method for assessment of structures formed by elongated grains using the Fourier transform technique: The image is transformed into reciprocal (Fourier) space and use is made of the fact that the Fourier transform is very sensitive to the orientation distribution of grains. The Fourier transform of the micrograph is a specific pattern, which bears a clear relation to the original micrograph and can be evaluated both qualitatively and quantitatively.
The task of pattern recognition can be greatly facilitated if position, rotation and size normali... more The task of pattern recognition can be greatly facilitated if position, rotation and size normalisation are incorporated within the pre-processing stage. Having obtained the required normalisation, then either feature extraction, template matching, artificial neural network systems or n-tuple logic node techniques may be used for object recognition. This paper outlines the general concepts of position, rotation and size normalisation followed by n-tuple and MIN/MAX node techniques for colour image recognition. The developed software system, which includes these normalisation and recognition methods, is briefly described. Finally, some results are presented which illustrate the performance obtainable from the developed software colour recognition system.
2012 International Conference on Applied Electronics, 2012
Based upon the n-tuple methodology originated by Bledsoe and Browning, this paper outlines the be... more Based upon the n-tuple methodology originated by Bledsoe and Browning, this paper outlines the benefits obtained by using grouped n-tuple nodes to enhance the pattern selectivity of single-layer n-tuple networks. The principles of the n-tuple and grouped node techniques are outlined and comparative results are presented. From the documented results it is confirmed that, compared with a conventional n-tuple system and with no higher memory requirements, the grouped node, Pseudo Huge Tuple (PHT) technique significantly improves the recognition confidence levels. Additionally, this technique tends to provide faster training and allows the user more flexibility in choosing system parameters.
This article describes a program simulation of neural networks. The program simulates the neural ... more This article describes a program simulation of neural networks. The program simulates the neural network based on Min/Max nodes. The simulation is used for the examination of the influence of interference. The recognition of analogous signals with superposed noise is tested. The recognition in time domain is used here. Potřeba rozpoznavani obrazů a signalů je v dnesni době cim dal castějsi v mnoha oborech lidske cinnosti. Proto je tato problematika stale předmětem intenzivniho vývoje. K tomuto ucelu je vyuivano mnoho různých algoritmů, zaloených předevsim na principu pouiti neuronových siti. V praxi jsou signaly, ktere chceme rozpoznavat, casto pod vlivem silneho ruseni. Proto je velmi důleita otazka odolnosti těchto algoritmů proti ruseni a jejich mone vyuiti v praxi pro potlaceni vlivu tohoto ruseni.
This article deals with a method of image recognition based on neural networks composed of MIN/MA... more This article deals with a method of image recognition based on neural networks composed of MIN/MAX nodes. The general concepts of the MIN/MAX nodes and the neural networks are outlined. The developed software system is then briefly introduced. Finally, the design of the neural networks in VHDL (VHSIC Hardware Description Language) is presented.
Drowsy driving is considered one of the most serious causes of fatal traffic accidents, especiall... more Drowsy driving is considered one of the most serious causes of fatal traffic accidents, especially for long distance drivers who struggle significantly during monotonous driving conditions. As a result, various types of warning systems have been proposed among other preventive measure against drowsy driving. These systems issue a warning upon detecting a decline in driving performance as indicated by a significant lane departure or fluctuation in headway distance, or upon detecting unsafe driver behavior such as prolonged eye closure time, or increased head sway, among other signs. Because these systems usually detect such behavior after drivers have already become drowsy, they may not provide sufficient active safety performance, and may not contribute significantly in decreasing related fatalities. Therefore, there is a need to pursue alternative ways for an earlier detection of drowsy driving. The main contribution of this work is the development of a new method for appropriate p...
This paper documents the results obtainable when either networks of n-tuple (NT) logic nodes or M... more This paper documents the results obtainable when either networks of n-tuple (NT) logic nodes or Min/Max (MM) nodes are used to recognize monochrome and color images. Networks of n-tuple nodes are based upon the n-tuple methodology originated by Bledsoe and Browning. Networks of Min/Max nodes employ similar techniques but are capable of directly processing grey-scale values. Both networks are illustrated with appropriate diagrams. The comparative results indicate that color recognition generally provides improved confidence levels. Also, overall, for `equivalent' networks there are insignificant differences amongst the responses. By means of implementing grouping of the nodes, higher confidence levels can be obtained. Of importance, if a suitable threshold is applied to the summed values of grouped nodes, the nets can be configured to provide confidence levels approaching 100%. The documented methodologies are relatively simple to implement in either hardware or software, easy to...
2019 International Conference on Applied Electronics (AE), Sep 1, 2019
Technology for modifying the local temperature of a material is tested for hot forming. A system ... more Technology for modifying the local temperature of a material is tested for hot forming. A system for selective controlled cooling was designed for this purpose and a workplace enabling heating, local cooling and material forming was built. This paper describes the principles and possibilities of the local cooling unit, control methods and measurement of the temperature fields.
Neural Network World, 2012
An image recognition system can be based on a single-layer neural network composed of Min/Max nod... more An image recognition system can be based on a single-layer neural network composed of Min/Max nodes. This principle is easy to use for greyscale images. However, this article deals with the possibilities of utilising neural nets for colour image recognition. Several principles are demonstrated and tested by recently developed software. A new modified Min/Max node Single Layer Net, suitable for recognition in HSV (Hue Saturation Value) colour space, is presented in this paper.
Journal of Applied Biomedicine
The aim of our work was to prepare part of the input data for a computational biomechanical model... more The aim of our work was to prepare part of the input data for a computational biomechanical model of both the active and passive elements of the tunica media of an aortic aneurysm. We analyzed tissue samples of the anterior wall of the normal, atherosclerotic and aneurysmatic subrenal abdominal aorta. We assessed the proportions of smooth muscle cells, elastin and collagen in histological sections of these samples and studied the morphological characteristics of the elastin network in the tunica media. Selected photomicrographs were studied, representing relatively well preserved areas without artifacts, ruptures, corrupted integrity of the tunica media or total elastinolysis. A new method was introduced for the assessment of structures formed by elastin membranes and fibres, using the fast Fourier transform (FFT) technique. The image was transformed into reciprocal (Fourier) space and the method made use of the fact that the FFT was very sensitive to the orientation distribution of thresholded elastin morphology. The results of this comparative study, obtained from selected samples from 24 patients, revealed that the percentage values of the constituents of the arterial wall can not distinguish between the preserved segments of normal, atherosclerotic or aneurysmatic aorta. The results of the Fourier analysis proved that the FFT provided an efficient method for evaluating cross sections of the elastin membranes and fibres, reflecting their anisotropy. The shape of the power spectrum of elastin was a simple pattern, whose description was quantified by the shape of its polar coordinates histogram. We discuss the methodological difficulties and biomechanical implications of our work as well compare it to other methods of elastin analysis.
2012 International Conference on Applied Electronics, Sep 1, 2012
This article describes a program simulation of neural networks. The program simulates the neural ... more This article describes a program simulation of neural networks. The program simulates the neural network based on Min/Max nodes. The simulation is used for the examination of the influence of interference. The recognition of analogous signals with superposed noise is tested. The recognition in time domain is used here.
2013 International Conference on Applied Electronics, Sep 1, 2013
EFTA 2003. 2003 IEEE Conference on Emerging Technologies and Factory Automation. Proceedings (Cat. No.03TH8696), 2003
Applied Electronics 2002, 2002
Oriented structures in metallography are usually evaluated by the intersection method or on the b... more Oriented structures in metallography are usually evaluated by the intersection method or on the base of comparison against a series of standard micrographs. In this paper is introduced a new method for assessment of struct ures formed by elongated grains using the Fourier transform technique: The image is transformed into reciprocal (Fourier) space and use is made of the fact that the Fourier transform is very sensitive to the orientation distribution of grains. The Fourier transform of the micrograph is a specific pattern, which bears a clear relation to the original micrograph and can be evaluated both qualitatively and quantitatively.
Elektrotechnika a informatika 2001, pp. 65-66, ISBN 80-7082-804-8, 2001
Design and implementation of neural network for signal recognition in VHDL language. This artic... more Design and implementation of neural network for signal recognition in VHDL language.
This article deals with a design of neural networks for signal recognition and system implementation in VHDL language. The neural networks are designed for the signal recognition and the suppression of the influence of interference. The networks are based on MIN/MAX nodes and a method of recognition in time domain is used.
Applied Electronics 2001, 2001
This paper outlines the general concepts of the use of networks of ‘trixel’ ‘n-tuple and ‘Min/Max... more This paper outlines the general concepts of the use of networks of ‘trixel’ ‘n-tuple and ‘Min/Max’node techniques for the recognition of coloured images.The developed software system, which also includes the options for position, size and rotation normalisation, is briefly described.
Finally, some results are presented which illustrate the performance obtainable from the developed software colour recognition system.
Journal of Real-Time Image Processing, 2013
This paper deals with the design and the implementation of an image recognition system based on F... more This paper deals with the design and the implementation of an image recognition system based on FPGA devices. It explores an n-tuple methodology using node ‘grouping’ and the possible advantages offered by this little-known technique. The paper is based on the implementation of this concept by an FPGA device. A novel approach to the organization of the neural networks data in the n-tuple memory is introduced. The system was tested on a real-world recognition task—the recognition of road signs. The test results are presented and discussed. It is concluded that the designed system may be a powerful part of more complex equipment for the solution of many recognition issues.
This article deals with a design of CAN bus analyser. For this purpose the design with utilisatio... more This article deals with a design of CAN bus analyser. For this purpose the design with utilisation of Field Programmable Gate Arrays (FPGAs) is used. A system structure consisted of basic blocks is introduced and the blocks are described. For the structure description is used VHDL (VHSIC Hardware Description Language).
Oriented structures in metallography are usually evaluated by the intersection method or on the b... more Oriented structures in metallography are usually evaluated by the intersection method or on the base of comparison against a series of standard micrographs. In this paper is introduced a new method for assessment of structures formed by elongated grains using the Fourier transform technique: The image is transformed into reciprocal (Fourier) space and use is made of the fact that the Fourier transform is very sensitive to the orientation distribution of grains. The Fourier transform of the micrograph is a specific pattern, which bears a clear relation to the original micrograph and can be evaluated both qualitatively and quantitatively.
The task of pattern recognition can be greatly facilitated if position, rotation and size normali... more The task of pattern recognition can be greatly facilitated if position, rotation and size normalisation are incorporated within the pre-processing stage. Having obtained the required normalisation, then either feature extraction, template matching, artificial neural network systems or n-tuple logic node techniques may be used for object recognition. This paper outlines the general concepts of position, rotation and size normalisation followed by n-tuple and MIN/MAX node techniques for colour image recognition. The developed software system, which includes these normalisation and recognition methods, is briefly described. Finally, some results are presented which illustrate the performance obtainable from the developed software colour recognition system.
2012 International Conference on Applied Electronics, 2012
Based upon the n-tuple methodology originated by Bledsoe and Browning, this paper outlines the be... more Based upon the n-tuple methodology originated by Bledsoe and Browning, this paper outlines the benefits obtained by using grouped n-tuple nodes to enhance the pattern selectivity of single-layer n-tuple networks. The principles of the n-tuple and grouped node techniques are outlined and comparative results are presented. From the documented results it is confirmed that, compared with a conventional n-tuple system and with no higher memory requirements, the grouped node, Pseudo Huge Tuple (PHT) technique significantly improves the recognition confidence levels. Additionally, this technique tends to provide faster training and allows the user more flexibility in choosing system parameters.
This article describes a program simulation of neural networks. The program simulates the neural ... more This article describes a program simulation of neural networks. The program simulates the neural network based on Min/Max nodes. The simulation is used for the examination of the influence of interference. The recognition of analogous signals with superposed noise is tested. The recognition in time domain is used here. Potřeba rozpoznavani obrazů a signalů je v dnesni době cim dal castějsi v mnoha oborech lidske cinnosti. Proto je tato problematika stale předmětem intenzivniho vývoje. K tomuto ucelu je vyuivano mnoho různých algoritmů, zaloených předevsim na principu pouiti neuronových siti. V praxi jsou signaly, ktere chceme rozpoznavat, casto pod vlivem silneho ruseni. Proto je velmi důleita otazka odolnosti těchto algoritmů proti ruseni a jejich mone vyuiti v praxi pro potlaceni vlivu tohoto ruseni.
This article deals with a method of image recognition based on neural networks composed of MIN/MA... more This article deals with a method of image recognition based on neural networks composed of MIN/MAX nodes. The general concepts of the MIN/MAX nodes and the neural networks are outlined. The developed software system is then briefly introduced. Finally, the design of the neural networks in VHDL (VHSIC Hardware Description Language) is presented.
Drowsy driving is considered one of the most serious causes of fatal traffic accidents, especiall... more Drowsy driving is considered one of the most serious causes of fatal traffic accidents, especially for long distance drivers who struggle significantly during monotonous driving conditions. As a result, various types of warning systems have been proposed among other preventive measure against drowsy driving. These systems issue a warning upon detecting a decline in driving performance as indicated by a significant lane departure or fluctuation in headway distance, or upon detecting unsafe driver behavior such as prolonged eye closure time, or increased head sway, among other signs. Because these systems usually detect such behavior after drivers have already become drowsy, they may not provide sufficient active safety performance, and may not contribute significantly in decreasing related fatalities. Therefore, there is a need to pursue alternative ways for an earlier detection of drowsy driving. The main contribution of this work is the development of a new method for appropriate p...
This paper documents the results obtainable when either networks of n-tuple (NT) logic nodes or M... more This paper documents the results obtainable when either networks of n-tuple (NT) logic nodes or Min/Max (MM) nodes are used to recognize monochrome and color images. Networks of n-tuple nodes are based upon the n-tuple methodology originated by Bledsoe and Browning. Networks of Min/Max nodes employ similar techniques but are capable of directly processing grey-scale values. Both networks are illustrated with appropriate diagrams. The comparative results indicate that color recognition generally provides improved confidence levels. Also, overall, for `equivalent' networks there are insignificant differences amongst the responses. By means of implementing grouping of the nodes, higher confidence levels can be obtained. Of importance, if a suitable threshold is applied to the summed values of grouped nodes, the nets can be configured to provide confidence levels approaching 100%. The documented methodologies are relatively simple to implement in either hardware or software, easy to...
2019 International Conference on Applied Electronics (AE), Sep 1, 2019
Technology for modifying the local temperature of a material is tested for hot forming. A system ... more Technology for modifying the local temperature of a material is tested for hot forming. A system for selective controlled cooling was designed for this purpose and a workplace enabling heating, local cooling and material forming was built. This paper describes the principles and possibilities of the local cooling unit, control methods and measurement of the temperature fields.
Neural Network World, 2012
An image recognition system can be based on a single-layer neural network composed of Min/Max nod... more An image recognition system can be based on a single-layer neural network composed of Min/Max nodes. This principle is easy to use for greyscale images. However, this article deals with the possibilities of utilising neural nets for colour image recognition. Several principles are demonstrated and tested by recently developed software. A new modified Min/Max node Single Layer Net, suitable for recognition in HSV (Hue Saturation Value) colour space, is presented in this paper.
Journal of Applied Biomedicine
The aim of our work was to prepare part of the input data for a computational biomechanical model... more The aim of our work was to prepare part of the input data for a computational biomechanical model of both the active and passive elements of the tunica media of an aortic aneurysm. We analyzed tissue samples of the anterior wall of the normal, atherosclerotic and aneurysmatic subrenal abdominal aorta. We assessed the proportions of smooth muscle cells, elastin and collagen in histological sections of these samples and studied the morphological characteristics of the elastin network in the tunica media. Selected photomicrographs were studied, representing relatively well preserved areas without artifacts, ruptures, corrupted integrity of the tunica media or total elastinolysis. A new method was introduced for the assessment of structures formed by elastin membranes and fibres, using the fast Fourier transform (FFT) technique. The image was transformed into reciprocal (Fourier) space and the method made use of the fact that the FFT was very sensitive to the orientation distribution of thresholded elastin morphology. The results of this comparative study, obtained from selected samples from 24 patients, revealed that the percentage values of the constituents of the arterial wall can not distinguish between the preserved segments of normal, atherosclerotic or aneurysmatic aorta. The results of the Fourier analysis proved that the FFT provided an efficient method for evaluating cross sections of the elastin membranes and fibres, reflecting their anisotropy. The shape of the power spectrum of elastin was a simple pattern, whose description was quantified by the shape of its polar coordinates histogram. We discuss the methodological difficulties and biomechanical implications of our work as well compare it to other methods of elastin analysis.
2012 International Conference on Applied Electronics, Sep 1, 2012
This article describes a program simulation of neural networks. The program simulates the neural ... more This article describes a program simulation of neural networks. The program simulates the neural network based on Min/Max nodes. The simulation is used for the examination of the influence of interference. The recognition of analogous signals with superposed noise is tested. The recognition in time domain is used here.
2013 International Conference on Applied Electronics, Sep 1, 2013
EFTA 2003. 2003 IEEE Conference on Emerging Technologies and Factory Automation. Proceedings (Cat. No.03TH8696), 2003
Applied Electronics 2002, 2002
Oriented structures in metallography are usually evaluated by the intersection method or on the b... more Oriented structures in metallography are usually evaluated by the intersection method or on the base of comparison against a series of standard micrographs. In this paper is introduced a new method for assessment of struct ures formed by elongated grains using the Fourier transform technique: The image is transformed into reciprocal (Fourier) space and use is made of the fact that the Fourier transform is very sensitive to the orientation distribution of grains. The Fourier transform of the micrograph is a specific pattern, which bears a clear relation to the original micrograph and can be evaluated both qualitatively and quantitatively.
Elektrotechnika a informatika 2001, pp. 65-66, ISBN 80-7082-804-8, 2001
Design and implementation of neural network for signal recognition in VHDL language. This artic... more Design and implementation of neural network for signal recognition in VHDL language.
This article deals with a design of neural networks for signal recognition and system implementation in VHDL language. The neural networks are designed for the signal recognition and the suppression of the influence of interference. The networks are based on MIN/MAX nodes and a method of recognition in time domain is used.
Applied Electronics 2001, 2001
This paper outlines the general concepts of the use of networks of ‘trixel’ ‘n-tuple and ‘Min/Max... more This paper outlines the general concepts of the use of networks of ‘trixel’ ‘n-tuple and ‘Min/Max’node techniques for the recognition of coloured images.The developed software system, which also includes the options for position, size and rotation normalisation, is briefly described.
Finally, some results are presented which illustrate the performance obtainable from the developed software colour recognition system.