Showing papers by "Sigeru Omatu published in 1994"

Neural network approach to land cover mapping

Abstract: A pattern classification method is proposed for remote sensing data using neural networks. First, the authors apply the error backpropagation (BP) algorithm to classify the remote sensing data. In this case, the classification performance depends on a training data set. In order to get stable and precise classification results, the training data set is selected based on geographical information and Kohonen's self-organizing feature map. Using the training data set and the error backpropagation algorithm, a layered neural network is trained such that the training patterns are classified with a specified accuracy. After training the neural network, some pixels are deleted from the original training data set if they are incorrectly classified and a new training data set is built up. Once training is complete, a testing data set is classified by using the trained neural network. The classification results of LANDSAT TM data show that this approach produces excellent results which are more realistic and noiseless compared with a conventional Bayesian method. >

Neural network approach to land cover mapping

Abstract: A pattern classification method is proposed for remote sensing data using neural networks. First, the authors apply the error backpropagation (BP) algorithm to classify the remote sensing data. In this case, the classification performance depends on a training data set. In order to get stable and precise classification results, the training data set is selected based on geographical information and Kohonen's self-organizing feature map. Using the training data set and the error backpropagation algorithm, a layered neural network is trained such that the training patterns are classified with a specified accuracy. After training the neural network, some pixels are deleted from the original training data set if they are incorrectly classified and a new training data set is built up. Once training is complete, a testing data set is classified by using the trained neural network. The classification results of LANDSAT TM data show that this approach produces excellent results which are more realistic and noiseless compared with a conventional Bayesian method. >

Pattern recognition apparatus and method of optimizing mask for pattern recognition according to genetic algorithm

Abstract: A bill-recognition apparatus includes a neural network having a learning capability and performs high-efficiency pattern recognition of seven kinds of U.S. dollar bills. Pattern image data optically inputted through a sensor is compressed using plurality of column masks, and then a plurality of values representative of images (slab values) are determined. The image data is divided into a large number of strip-shaped segments, and some of theses segments are masked with column areas of masks. The values representative of images compressed through column masks are not influenced by a slight inclination of the pattern image during the reading operation. These values representative of images are inputted to a separation processing unit (neural network). From these values, the separation processing unit calculates separation values corresponding to respective decision patterns associated with pattern images, using weights which have been adjusted to optimum values for respective decision patterns. A correct pattern image is determined from the maximum value of the separation values. The above arrangement allows for a reduction in scale of the neural network and control system. Furthermore, bill recognition may also be achieved by separation processing using a plurality of small-scaled neural networks connected in cascade, or replacing weight functions in the same neural network and performing separation processing a plurality of times for the same slab values (cascade processing). In this way, it is possible to reduce the scale of the neural network and the control system.

A paper currency recognition method by a small size neural network with optimized masks by GA

Abstract: Compactness, transaction speed, and cost are important design factors when we apply neural networks to commercial products. We propose a structure reduction method for NNs. We adopt slab values which are sums of input pixels as characteristics of the inputs. But there is the possibility of generating the same slab values even when the inputs are different. To avoid this problem, we adopt a mask which covers some parts of the input. This enables us to reflect the difference of input pattern to slab values with masks. Furthermore, we adopt the genetic algorithm (GA) to optimize the masks. We can generate various effective masks automatically. Finally, we show that the proposed method by neuro-recognition with masks can be applied effectively to paper currency recognition machine using the GA. >

A Neuro-Money Recognition Using Optimised Masks by GA

Abstract: Up to now, much research of the application to neural networks (NN) has been reported We have proposed a neuro-pattern recognition for bill money with masks and have reported its effectiveness for money recognition Recently, genetic algorithm (GA) is reported as the effective optimizing method In this paper, we adopt the GA to mask optimization in the recognition method Namely, we regard the position of the masked part in the input image as a gene We operate crossover, selection, and mutation to some genes By repeating a series of these operations, we can get effective masks for paper currency recognition We compare the ability of NN using the optimized masks by the GA with the one of NN using the random masks determined by random numbers Then we show that the GA is effective to optimize masks for the method of neuro-pattern recognition with masks Furthermore, we develop high-speed neuro-recognition board to realize the neuro-pattern recognition for paper currency in the commercial products

Improvement of speed control performance using PID type neurocontroller in an electric vehicle system

Abstract: In order to develop an efficient driving system for electric vehicle (EV), a testing system using motors has been built to simulate the driving performance of EVs. In the testing system, the PID controller is used to control rotating speed of motor when the EV drives. In this paper, in order to improve the performance of speed control, a neural network is applied to tuning parameters of PI controller. It is shown,through experiments that a neural network can reduce output error effectively while the PI controller parameters are being tuned online. >

Classification of elongated and contracted images using new regular moments

Abstract: This paper presents a technique to classify images that have been elongated or contracted. The problem is formulated using conventional regular moments. It is shown that the conventional regular moment-invariants remain no longer invariant when the image is scaled unequally in the x- and y-directions. A method is proposed to form moment-invariants that do not change under such unequal scaling. Results of computer simulations for images are also included verifying the validity of the method proposed. >

Stabilization of inverted pendulum by neuro-control

Abstract: In this paper, we consider stabilization of an inverted pendulum which can be controlled by moving a cart in an intelligent way. Here we adopt a fuzzy control to swing up the pendulum from the bottom and switch the control law from fuzzy control to linear quadratic optimal control after the pendulum approaches to an upper position. But there is some fluctuation or offset in a position even if we use the linear quadratic control law. In order to remove these phenomena, we use a neuro-controller. The experimental results show the effectiveness of the present approach. >

Rotation invariant neural pattern recognition system which can estimate a rotation angle

Abstract: This paper presents a rotation invariant neural pattern recognition system, which can recognize a rotated pattern and estimate a rotation angle. The system is very effective for a rotated coin recognition problem, but is poor compared with human performance. It is well known that human sometimes recognizes a rotated pattern by means of the mental rotation. Such a fact, however, has never been considered and used in neural pattern recognition systems, especially in rotation invariant systems. Therefore, we examine the principle of mental rotation and apply it to a rotation invariant pattern recognition system. The system with such a principle could recognize a rotated pattern and estimate a rotation angle. It is shown that the system is effective to recognize a rotated pattern from results of computer simulation for a coin recognition problem. >

Neuro-pattern classification using Zernike moments and its reduced set of features

Abstract: The paper proposes a neural network technique to classify numerals using Zernike moments that are invariant to rotation only. In order to make them invariant to scale and shift, we introduce modified Zernike moments based on regular moments. Owing to the large number of Zernike moments used, it is computationally more efficient to select a subset of them that can discriminate as well as the original set. The subset is determined using stepwise discriminant analysis. The performance of a subset is examined through its comparison to the original set. The results are shown of using such a scheme to classify scaled, rotated, and shifted binary images and images that have been perturbed with random noise. In addition to the neural network approach, the Fisher's classifier is also used, which is a parametric classifier. A comparative study of their performances shows that the neural network approach produces better classification accuracy than the Fisher's classifier. When a suitable subset of Zernike moments is used, the classifiers perform well, just like the original set. The performance of the classifiers is also examined. The computational time is greatly reduced when a suitable subset of Zernike moments is used.