Keisuke Kameyama

Semiconductor defect classification using hyperellipsoid clustering neural networks and model switching

An automatic defect classification (ADC) system for visual inspection of semiconductor wafers, using a neural network classifier is introduced The proposed hyperellipsoid clustering network (HCN) employing a radial basis function (RBF) in the hidden layer is trained with additional penalty conditions for recognizing unfamiliar inputs as originating from an unknown defect class. Also, by using a dynamic model alteration method called model switching, a reduced-model classifier which enables an efficient classification is obtained In the experiments, the effectiveness of the unfamiliar input recognition was confirmed, and a classification rate sufficiently high for use in the semiconductor fab was obtained.

Estimation of the stress distribution in metal using nonlinear acoustoelasticity

A new nondestructive method to estimate the stress distribution in a metal is proposed. It is based on the use of nonlinear dependence of ultrasound velocity on stress in metals. First, the dependence of ultrasound velocity on stress in an aluminum alloy is observed, and it is confirmed that the relation is nonlinear and that its derivative changes monotonically in the stress range of interest. Hence, if a certain stress perturbation is given to the observing region in a metal and the change of ultrasound velocity due to the perturbation is measured, the stress in that region can be estimated. This is the basic idea of the measurement of stress in a metal. Now, the field to give the stress perturbation is scanned along the probing ultrasonic beam and a set of data of phase change due to the velocity change is acquired. Then a matrix inversion technique is applied to them by taking account of the spatial distribution of the stress perturbation to obtain a precise stress distribution along the probing beam....

Imaging System of Precise Hardness Distribution in Soft Tissue in Vivo Using Forced Vibration and Ultrasonic Detection

This paper focuses on observation of biological tissue hardness (elasticity) by applying low frequency vibration to the tissue and detecting the vibration propagation characteristics with use of ultrasonic pulsed Doppler technique. It has been reported that the amplitude, phase and velocity of the propagating vibration can be mapped using ultrasonic Doppler method [1]. Traditionally, low frequency vibration propagation in soft tissues have been considered to be shear waves, and therefore tissue hardness have been formulated to be directly related to the propagating velocity of the vibration [2]. However, through experimental verification, it was found that vibration propagation is much affected by the vibrating conditions such as vibration frequency or the shape of the vibrator attachment. Here, a new formulation of vibration propagation in soft tissue which takes the vibrating conditions into consideration is introduced. Based on the derived formulation, a precise tissue hardness map estimation which is independent of the vibrating conditions is demonstrated. Additionally, a method to scan the vibration frequency and to map the average hardness estimated for multiple frequencies is used in order to reduce the effects of the developed standing waves which can be a serious obstacle for estimation of precise hardness maps. The method proved to be useful for reducing the false effects (ghosts) observed in hardness maps. The feasibility of the introduced methods were certified through observation of hardness for an agarose phantom with different elasticity and in vivo human thigh with various loads at the ankle.

A neural network incorporating adaptive Gabor filters for image texture classification

A novel neural network architecture for image texture classification is introduced. The proposed kernel modifying neural network (KM Net), which incorporates a convolution filter kernel array and a classifier in one, enables an automated texture feature extraction in the multichannel texture classification through modification of the kernels and the connection weights by a backpropagation-based training rule. The first layer units working as the convolution kernels are constrained to be an array of Gabor filters, which achieves the most efficient texture feature localization. The following layers work as a classifier of the extracted texture feature vectors. The capability of the KM Net and its training rule is verified with basic problems of synthetic and fabric texture images, and also with a biological tissue classification problem in an ultrasonic echo image.

CCE-based index selection for neuro assisted MR-image segmentation

For image segmentation with the aid of neural networks of a reasonable size, it is important to select the most effective combination of secondary indices to be used for the classification. Here, the authors introduce a vector quantized conditional class entropy (VQCCE) criterion to evaluate which indices are effective for pattern classification, without testing on the actual classifiers. The proposed method was successfully applied for brain MR segmentation problems to classify the gray-matter/white-matter regions.

Model switching by channel fusion for network pruning and efficient feature extraction

Introduces a feature dimension reduction method called channel fusion, and a criterion for redundant channel detection called effective map distance. Channel fusion locally reduces the feature dimension by replacing the redundant channel pair with a single channel, suppressing the map distance between the two models. It is applicable to network model switching such as pruning hidden layer units and reducing input channels. Effective map distance is a measure of discrepancy in the models before and after the channel reduction, which can be defined for any dimension reduction strategy. The two methods were applied to the feature extraction layer of a network for image texture classification. Improvements both in the classification rate and the training speed were observed when the methods were used during the training, which dynamically enabled us to switch the model for efficient feature extraction.

Acoustical tissue nonlinearity characterization using bispectral analysis

Abstract A novel method for measuring the biological soft tissue nonlinearity from nonlinear propagation characteristics of low frequency vibration using bispectral analysis is proposed. The nonlinearity of the biological tissue is represented by an equivalent second order parameter of nonlinearity Γ. By formulating the waveform distortion due to medium nonlinearity, this Γ is represented as a function of the wave distortion ( N ) defined as the amplitude ratio of the third harmonics to the fundamental frequency components. In order for adequate estimation of N from the noisy vibrational waveform detected by ultrasonic pulse Doppler method, bispectral analysis which is immune to additive Gaussian noise was used. Measurement results of Γ for pig tissue (pork) and human tissue both in vitro and in vivo will be shown.

Simultaneous Imaging System of Three Kinds of Parameters of Nonlinearity for Medical Diagnosis

This paper discusses an imaging system for biological tissue characterization for medical diagnosis. This system uses ultrasonic wave (5MHz) for probing and pump wave (350kHz) for pressure perturbation (~latrn). In addition to the most widely used pulsed echo image (linear reflectivity), on this system, we suggest three kinds of tissue characteristic parameters which are obtained by an interaction of the tissue and the pump wave as listed in the following. n n1. n n1. Probe phase shift due to sound velocity dependency to pressure. (Phase Shift Parameter) n n n n n2. n nProbe phase distortion caused by microstructural positional movability. n n n n n3. n nReflectivity change caused by microstructural change in orientation due to pumping.

Robot vision system using high order correlation analysis

Object movement detection by high order correlation analysis of optical sensor array signals is introduced. The optical sensors observe the moving object surface which is assumed to be a non-uniform speckle-like texture. The measurement system is applicable to general robotic movement detection because: it employs a noncontact measurement method, the system can be made very compact, and it enables approximation of the movement trace with a sequence of arcs instead of the conventional connection of simple line segments. The authors looked into estimation of the running trace of an autonomous vehicle by observing the ground pattern.<<ETX>>

Bispectral analysis applied for measurement of nonlinear characteristics of vibration propagation in soft tissues

A new method of measurement of nonlinear propagation characteristics of low frequency vibration in the tissues by using bispectral analysis is proposed. The main nonlinear characteristic of low frequency vibration in soft tissue can be represented as amplitude ratio of 3rd harmonics to the fundamental frequency components defined by the parameter N. The evolution of N as function of amplitude of low frequency vibration at an observing point in the interested region of tissue is observed, and the parameter of nonlinearity of medium is estimated based on the formulation. For measurement of N the bispectral analysis was adopted so that signal to noise ratio is increased in connection with the ultrasonic pulse Doppler method. Experimental results for agar phantom and pig tissue (pork) are shown.<<ETX>>