Shigeru Omatu

Early stage fire detection using reliable metal oxide gas sensors and artificial neural networks

Conventional fire detectors use the smoke density or the high air temperature to trigger the fire alarm. These devices lack of ability to detect the source of fire in the early stage and they always create false alarms. In this paper, a reliable electronic nose (EN) system designed from the combination of various metal oxide gas sensors (MOGS) is applied to detect the early stage of fire from various sources. The time series signals of the same source of fire in every repetition data are highly correlated and each source of fire has a unique pattern of time series data. Therefore, the error backpropagation (BP) method can classify the tested smell with 99.6% of correct classification by using only a single training data from each source of fire. The results of the k-means algorithms can be achieved 98.3% of correct classification which also show the high ability of the EN to detect the early stage of fire from various sources accurately.

Italian Lira classification by LVQ

In this paper, a new method to classify the Italian Liras by using the learning vector quantization (LVQ) is proposed. The Italian Liras of 8 kinds, 1000, 2000, 5000, 10000, 50000 (new), 50000 (old), 100000 (new), 100000 (old) Liras with four directions A,B,C, and D are used, where A and B mean the normal direction and the upside down direction and C and D mean the reverse version of A and B. The original image with 128 by 64 pixels is observed at the transaction machine in which rotation and shift are included. After correction of these effects, we select a suitable area which shows the bill image and feed the image with 64 by 15 pixels to a neural network. Although the neural network of the LVQ type can process in any order of the dimension of the input data, the smaller size is better to achieve a faster convergence.

Neuro-PID control for stabilization of inverted single and double pendulums

The paper is concerned with an architecture of a self-tuning neuro-PID control system and its application to stabilization of an inverted pendulum. A single-input multi-output system is considered to control the inverted pendulum by using the PID controller. The PID gains are tuned by using two kinds of neural networks. The simulation results show the effectiveness of the proposed approach.

An analysis of expression data using a support vector machine and dimensional reduction methods

Gene expression data are expected to be a significant aid in the development of efficient cancer diagnosis and classification platforms. However, gene expression data are high-dimensional and the number of samples is small in comparison to the dimensions of the data. Furthermore, the data are inherently noisy. Therefore, in order to improve the accuracy of the classifiers, we would be better off reducing the dimensionality of the data. As a method of dimensionality reduction, there are two previous proposals: feature selection and dimensionality reduction. Feature selection is a feedback method which incorporate the classifier algorithm in the future selection process. Dimensionality reduction refers to algorithms and techniques which create new attributes as combinations of the original attributes in order to reduce the dimensionality of a data set. In this article, we compared the feature selection methods and the dimensionality reduction methods, and verified the effectiveness of both types. For the feature selection methods we used one previously known method and three proposed methods, and for the dimensionality reduction methods we used one previously known method and one proposed method. From an experiment using a benchmark data set, we confirmed the effectiveness of our proposed method with each type of dimensional reduction method.

Fire detection systems by compact electronic nose systems using metal oxide gas sensors

In this paper, a reliable electronic nose (EN) system designed from the combination of various metal oxide gas sensors (MOGS) is applied to detect the early stage of fire from various sources. The time series signals of the same source of fire in every repetition data are highly correlated and each source of fire has a unique pattern of time series data. Therefore, the error backpropagation (BP) method can classify the tested smell with 99.6% of correct classification by using only a single training data from each source of fire. The results of the k-means algorithms can be achieved 98.3% of correct classification which also show the high ability of the EN to detect the early stage of fire from various sources accurately.

Monitoring rice-crop growth with ENVISAT-1/ASAR alternating polarization mode

The Hiroshima Institute of Technology (HIT) manages direct downlinks for microwave and optical earth observation satellite data in Japan. This study focuses on validating rice monitoring using ground truth data from ENIVISAT-1/ASAR, such as the height of rice crop, vegetation cover, and leaf area index in test sites in the Hiroshima district in Japan. ENVISAT-1/ASAR data can monitor the rice-crop growing cycle using alternating polarization (AP) mode images. However, ASAR data is influenced by several parameters, such as land-cover structure, and the direction and alignment of rice fields in the test sites. To investigate these parameters, in this study the validation was combined with microwave image data and ground truth data for rice-crop fields. Multitemporal, multidirection (descending and ascending), and multiangle ASAR AP-mode images were used to investigate the rice-crop growing cycle. Finally, the extraction of rice-planted areas was attempted using multitemporal ASAR AP mode data, such as VV/VH and HH/HV. This study clarifies that the estimated rice-planted area agrees with the existing statistical data for areas within the rice field. In addition, HH/HV is more effective than VV/VH in extracting the rice-planted area.

Failure analysis of transmission devices using self-organizing map

In this paper, a general failure analysis method of transmission devices is proposed. First, we record the acoustic signals of good and no good transmission devices in operation. Then we decompose the acoustic signals by using wavelet transform. From all the components of signal, we select the significant component, which corresponds to the specified failure, based on the reconstructed signal, while setting the lower level wavelet approximation to zero. Next, we compute the frequency characteristic of the significant component and use the self-organizing map (SOM) to classify the specified no good products from the good and other no good products. By comparing the difference between the groups of specified no good products and good products, we can estimate the cause of trouble. Furthermore, while an unknown product is inputted to the SOM network, we can determine whether the inputted product has the specified failure. The experimental results show that the proposed method can perform the failure analysis of transmission devices successfully.

Genetic relaxed iterative Fourier transform algorithm for image reconstruction from Fourier transform magnitude

This paper proposes an algorithm for image reconstruction from the Fourier transform magnitude by means of a genetic algorithm. Since the iterative Fourier transform algorithm involves repetitive application of constraints in the image domain and Fourier domain, correct image reconstruction is not necessarily achieved, and incorrect images may be obtained at certain initial settings (stagnation phenomenon). This study offers a solution to this stagnation problem, specifically, the genetic relaxed iterative Fourier transform algorithm. In it, genetic operations (selection, crossover, and mutation) are employed to avoid stagnation. Computer simulations show that the proposed algorithm is free of stagnation: that is, correct image reconstruction is obtained irrespective of the initial settings. In addition, the proposed algorithm proves faster than the conventional iterative Fourier transform algorithm, and the initial settings are easy to choose because the parameters do not depend on the image features. In this study, binary images are considered.