Kurosh Madani

Hybrid Salient Object Extraction Approach with Automatic Estimation of Visual Attention Scale

In this work we present an intelligent approach to detection and extraction of salient objects. The described system is inspired by early processing stages of human visual system and is based on our previous work on the field of visual saliency. Building on our preceding system, which worked with a fixed visual attention scale, we develop a machine learning approach using an artificial neural network and genetic algorithm, estimating automatically the visual attention scale for each input image individually. The whole approach has low complexity and can be run in speed close to real-time on contemporary processors. Quantitative evaluation results of the described approach with visual attention scale estimation are compared to results obtained with a fixed scale and to results of two other existing salient object detection techniques. The system is a part of our work on an intelligent machine vision system, using visual saliency for unsupervised learning of objects.

Data driven multiple neural network models generator based on a tree-like scheduler

In this paper we describe a new penalty-based model selection criterion for nonlinear models which is based on the influence of the noise in the fitting. According to Occams razor we should seek simpler models over complex ones and optimize the trade-off between model complexity and the accuracy of a models description to the training data. An empirical derivation is developed and computer simulations for multilayer perceptron with weight decay regularization are made in order to show the efficiency and robustness of the method in comparison with other well-known criteria for nonlinear systems.

Hybrid credit ranking intelligent system using expert system and artificial neural networks

The main goal of all commercial banks is to collect the savings of legal and real persons and allocate them as credit to industrial, services and production companies. Non repayment of such credits cause many problems to the banks such as incapability to repay the central bank’s loans, increasing the amount of credit allocations comparing to credit repayment and incapability to allocate more credits to customers. The importance of credit allocation in banking industry and it’s important role in economic growth and employment creation leads the development of many models to evaluate the credit risk of applicants. But many of these models are classic and are incapable to do credit evaluation completely and efficiently. Therefore the demand to use artificial intelligence in this field has grown up. In this paper after providing appropriate credit ranking model and collecting expert’s knowledge, we design a hybrid intelligent system for credit ranking using reasoning-transformational models. Expert system as symbolic module and artificial neural network as non-symbolic module are components of this hybrid system. Such models provide the unique features of each components, the reasoning and explanation of expert system and the generalization and adaptability of artificial neural networks. The results of this system demonstrate hybrid intelligence system is more accurate and powerful in credit ranking comparing to expert systems and traditional banking models.

Image Processing Using RBF like Neural Networks: A ZISC-036 Based Fully Parallel Implementation Solving Real World and Real Complexity Industrial Problems

The present article concerns neural based image processing and solutions developed for industrial problems using the ZISC-036 neuro-processor, an IBM hardware processor which implements the Restricted Coulomb Energy algorithm (RCE) and the K-Nearest Neighbor algorithm (KNN). The developed neural based techniques have been applied for image enhancement in order to restore old movies (noise reduction, focus correction, etc.), to improve digital television images, or to treat images which require adaptive processing (medical images, spatial images, special effects, etc.). We also have developed and implemented on ZISC-036 neuro-processor, a neural network based solution for visual probe mark inspection in VLSI production for the IBM Essonnes plant. The main characteristics of such systems are real-time control and high reliability in detection and classification tasks. Experimental results, validating presented concepts, have been reported showing quantitative and qualitative improvement as well as the efficiency our solutions.

A survey of artificial neural networks based fault detection and fault diagnosis techniques

We present several neural network based approaches to analog circuits fault defection and fault diagnosis using different neural networks based structures. The paper focuses on the one hand, on multi-neural networks approaches, and on the other hand, on real time implementation of neural based fault diagnosis techniques. Simulation and experimental results, are reported.

Visual Probe Mark Inspection, Using Hardware Implementation of Artificial Neural Networks, in VLSI Production

As a result of their adaptability, artificial neural networks present good solutions for a permanently increasing range of industrials problems. So, if their usefulness has already been confirmed, very few papers deal with real applications of this kind of technology. Our goal is to present a neural based solution that we have developed for visual inspection in VLSI production for the IBM Essonnes plant. The main characteristics of such systems are real-time control and high reliability in detection and classification tasks. The presented system is based on a ZISC©, an IBM hardware implementation of the Restricted Coulomb Energy algorithm and of the K-Nearest Neighbor algorithm. The goal of the developed application is to inspect vias for probe damage during wafer tests: each via is analyzed and classified (good impact, bad impact or absence of impact). First results are really encouraging and show the efficiency of this system in manufacturing environment.

Non Linear Process Identification Using a Neural Network Based Multiple Models Generator

Identification of non-linear systems is an important task for model based control, system design, simulation, prediction and fault diagnosis. In real world applications, strong linearity and large number of related parameters make the realization of those steps challenging, and so, the identification task difficult. Recently, a number of works based on Multiple Modelling have been proposed to avoid difficulties related to non-linearity. In this paper we use an Artificial Neural Network based data driven Multiple Model generator, that we called T-DTS (Treelike Divide To Simplify), for non-linear systems identification. T-DTS reduces modeling complexity on both data and processing levels. The efficiency of such approach has been analyzed trough two applications dealing with none-linear process identification. Experimental results validating our approach have been reported.

Neuro-vector-based electrical machine driver combining a neural plant identifier and a conventional vector controller

One of the most important problems, for a machine control process is the system identification. To identify varying parameters which are dependent from other systems parameters (speed, voltage and currents, etc.), one must have an adaptive control system. Synchronous machines conventional vector controls implementation using PID controllers have been recently proposed presenting the best actual solution. It supposes an appropriated model of the plant. But real plants parameters vary and the P.I.D. controller is not suitable because of the parameters variation and non-linearity introduced by the machines physical structure. In this paper, we present an on-line dynamic adaptive neural based vector control system identifying the motors parameters of a synchronous machine. We present and discuss a DSP based real- time implementation of our adaptive neuro-controller. Simulation and experimental results validating our approach have been reported.

Hybrid neural-based decision level fusion architecture: application to road traffic collision avoidance

A hybrid decision level architecture for a road collision risks avoidance system is presented. The goal of the decision level is to clas- sify the behavior of the vehicles observed by a smart system or vehicle. The knowledge of vehicle behavior enables the best management of the smart system resources. The association of a model to each observed vehicle mainly enables the limitation of inference and of the set of actions to be activated; thus the interactions between system levels can be more intelligent. The decision level of this architecture is composed of a neural classifier, which is associated to a numerical classifier. Each of these classifiers provides decisions that are expressed within the framework of fuzzy theory. An optimal fusion policy is reached using the functional neural network tool.

A cognitive approach for robots' vision using unsupervised learning and visual saliency

In this work we contribute to development of an online unsupervised technique allowing learning of objects from unlabeled images and their detection when seen again. We were inspired by early processing stages of human visual system and by existing work on human infants learning. We suggest a novel fast algorithm for detection of visually salient objects, which is employed to extract objects of interest from images for learning. We demonstrate how this can be used in along with state-of-the-art object recognition algorithms such as SURF and Viola-Jones framework to enable a machine to learn to re-detect previously seen objects in new conditions. We provide results of experiments done on a mobile robot in common office environment with multiple every-day objects.