Simon Y. Foo

Neural network-based EKG pattern recognition

Abstract The highly nonlinear chaotic nature of electrocardiogram (EKG) data represents a well-suited application of artificial neural networks (ANNs) for the detection of normal and abnormal heartbeats. Digitized EKG data were applied to a two-layer feed-forward neural network trained to distinguish between different types of heartbeat patterns. The Levenberg–Marquardt training algorithm was found to provide the best training results. In our study, the trained ANN correctly distinguished between normal heartbeats and premature ventricular contractions in 92% of the cases presented.

Analog components for the VLSI of neural networks

Principles of operation and basic building blocks of artificial neural networks are described. Deterministic components, comprising variable linear conductance devices and components used for processing elements, are discussed. These devices are analyzed using SPICE. The reasons for using simple analog circuits rather than digital circuits are examined.<<ETX>>

A link availability-based QoS-aware routing protocol for mobile ad hoc sensor networks

With more and more wireless devices being mobile, there is a constant challenge to provide reliable and high quality communication services among these devices. In this paper, we propose a link availability-based QoS-aware (LABQ) routing protocol for mobile ad hoc networks based on mobility prediction and link quality measurement, in addition to energy consumption estimate. The goal is to provide highly reliable and better communication links with energy-efficiency. The proposed routing algorithm has been verified by NS-2 simulations. The results have shown that LABQ outperforms existing algorithms by significantly reducing link breakages and thereby reducing the overheads in reconnection and retransmission. It also reduces the average end-to-end delay for data transfer and enhances the lifetime of nodes by making energy-efficient routing decisions.

A Stochastic Digital Implementation of a Neural Network Controller for Small Wind Turbine Systems

This letter presents a reconfigurable hardware implementation of feed-forward neural networks using stochastic techniques. The design is based on the stochastic computation theory to approximate the nonlinear sigmoid activation functions with reduced digital logic resources. The large parallel neural network structure is then implemented on a reconfigurable field-programmable gate array (FPGA) device with high fault tolerance capability. The method is applied to a neural-network based wind-speed sensorless control of a small wind turbine system. The experimental results confirmed the validity of the developed stochastic FPGA implementation. The general design method can be extended to include other power electronics applications with different feed-forward neural network structures

Hand pose estimation for American sign language recognition

In the foreseeable future, gestured inputs will be widely used in human-computer interfaces. This paper describes our initial attempt at recognizing 2D hand poses for application in video-based human-computer interfaces. Specifically, this research focuses on 2-D image recognition utilizing an evolved wavelet-based feature vector. We have developed a two layer feed-forward neural network that recognizes the 24 static letters in the American sign language (ASL) alphabet using still input images. Thus far, two wavelet-based decomposition methods have been used. The first produces an 8-element real-valued feature vector and the second a 18-element feature vector. Each set of feature vectors is used to train a feed-forward neural network using Levenberg-Marquardt training. The system is capable of recognizing instances of static ASL fingerspelling with 99.9% accuracy with an SNR as low as 2. We conclude by describing issues to be resolved before expanding the corpus of ASL signs to be recognized.

A fuzzy logic approach to fire detection in aircraft dry bays and engine compartments

In this paper, a fuzzy logic approach is applied to detect hydrocarbon fires in aircraft dry bays and engine compartments. The inputs to the fuzzy system consist of a set of statistical measures derived from the histogram and image subtraction analyses of successive image frames. Specifically, fuzzy rules based on the median, standard deviation, and normalized first-order moment statistical measures of histogram data and the mean statistical measure of image subtraction data of successive frames are used to compute the probability of a fire event. This fuzzy logic approach is also tested for false alarms such as those due to flashlights and high-power halogen lights. It is shown that image subtraction analysis can be used to accurately distinguish fires from false alarms.

Local exponential stability of competitive neural networks with different time scales

Abstract This contribution presents a new method of analyzing the dynamics of a biological relevant neural network with different time scales based on the theory of flow invariance. We are able to show that the resulting stability conditions are less restrictive and more general than with K -monotone theory or singular perturbation theory. The theoretical results are further substantiated by simulation results conducted for analysis and design of these neural networks.

A rule-based machine vision system for fire detection in aircraft dry bays and engine compartments

In this paper, a rule-based machine vision approach is applied to detect and categorize hydrocarbon fires in aircraft dry bays and engine compartments. Images for computer analysis are provided by charge-coupled device imaging sensors placed inside dry bays and engine compartments. Using a set of heuristics based on statistical measures derived from the histogram and image subtraction analyses of successive image frames, we showed that it is possible to detect and categorize life-threatening fires from non-fire/non-lethal events accurately in sub-millisecond response time. Specifically, the median, standard deviation, and first-order moment statistical measures of the histogram data of each image frame are used to confirm the presence or absence of fire. Concurrently, another set of mean, median, and standard deviation statistical measures from the image subtraction of two successive frames are used to determine the growth and subsequently reaffirm the existence of a fire. This approach is also tested for false alarms such as those due to flashlights and high-power halogen lights.

Model-free visualization of suspicious lesions in breast MRI based on supervised and unsupervised learning

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has become an important tool in breast cancer diagnosis, but evaluation of multitemporal 3D image data holds new challenges for human observers. To aid the image analysis process, we apply supervised and unsupervised pattern recognition techniques for computing enhanced visualizations of suspicious lesions in breast MRI data. These techniques represent an important component of future sophisticated computer-aided diagnosis (CAD) systems and support the visual exploration of spatial and temporal features of DCE-MRI data stemming from patients with confirmed lesion diagnosis. By taking into account the heterogeneity of cancerous tissue, these techniques reveal signals with malignant, benign and normal kinetics. They also provide a regional subclassification of pathological breast tissue, which is the basis for pseudo-color presentations of the image data. Intelligent medical systems are expected to have substantial implications in healthcare politics by contributing to the diagnosis of indeterminate breast lesions by non-invasive imaging.

Job-shop scheduling based on modified tank-hopfield linear programming networks

Abstract The Tank-Hopfield linear programming network is modified to solve job-shop scheduling, a classical optimization problem. Using a linear energy function, the approach described in this paper avoids the traditional problems associated with most Hopfield networks using quadratic energy functions. Although this approach requires more hardware (in terms of processing elements and resistive interconnects) than a recent approach by Zhou et al. (IEEE Trans. Neural Networks 2, 175–179, 1991) the neurons in the modified Tank-Hopfield network do not perform extensive calculations, unlike those described by Zhou et al.