Shahed Shahir

Fuzzy Associative Database for multiple planar object recognition

In this paper a comprehensive method for multiple planar object recognition is presented. For this purpose, a Fuzzy Associative Database (FAD) is developed. FAD consists of a Fuzzy Database (FD) and a Fuzzy Search Engine (FSE). FD holds the trained object information, as in human memory, and FSE performs as in brain, which processes incoming information based on information exists in memory, database. The FD includes two tables. The FSE uses table one to construct a Bank of Fuzzy Associative Memory Matrix (BFAMM) in order to conduct search over table two. In fact, the FSE establishes a correspondence between an object and one of the trained classes in table two of the FD.

PERMITTIVITY PROFILE ESTIMATION BASED ON NON-RADIATING EQUIVALENT SOURCE (2D CASE)

This paper presents a new approach to the electromagnetic inverse scattering formulation of the permittivity proflle estimation. The proposed approach is particularly efiective for the cases where unknown objects are made of a flnite number of homogeneous regions. This approach prevents the need for the Born approximation initial guess and updating the internal total electric fleld iteratively. The solution to the inverse source problem and scattering problem is not unique. To address the non-uniqueness issue, we have deflned the non-radiating objective functions. By minimizing this objective function and applying some constraints, we have been able to obtain a unique permittivity proflle. The simulation results indicate that the low-contrast and high-contrast permittivity proflles are accurately estimated by the proposed method. The distinguishing feature of the proposed approach is that by including the non-radiating part of the equivalent source, the unknown permittivity proflle becomes the solution to a minimization problem, which is much less computationally intensive as compared to existing methods using iterative fleld calculation over the entire domain, when applied to large (in terms of wavelength) objects. The high performance of the proposed method for noisy measured data has also been verifled.

Stand-alone vision sensor design based on fuzzy associative database

In this paper a comprehensive design methodology for stand-alone vision sensor with application in manufacturing production lines is presented. The sensor can be employed as a substitute for traditional sensors, such as photoelectric and proximity sensors. The advantage of the vision sensor over traditional manufacturing sensors is distinguishing the incoming product prior to determination of the product position. The sensor is designed based on Fuzzy Associative Database (FAD). The proposed method is extremely fast and suitable for real time processing. After training, the sensor will be able to recognize the coming product by conducting fuzzy search over the knowledge base. If the vision sensor recognized the product, the sensor provides the product category, position and speed for programmable logic controllers (PLC) or other controllers to perform further action.

Adaptive fuzzy associative memory for online quality control

In this paper, an online quality inspection is presented based on the adaptive fuzzy associative memory (AFAM) theory. The AFAM along with vision technology enables us to inspect the quality of each component online. Throughout the process, four different types of classification exist, namely, desired, stretched, squeezed and deformed foam barrier. The learning vector quantization (LVQ) is applied to train the system based on the defined clusters according to the trainees. After ending a course of training, a bank of fuzzy associative memory (BFAM) is constructed. To perform online quality inspection, the composition applies to the input fuzzy vector and BFAM.

Monte Carlo based non-radiating objective function minimization for permittivity profile estimation

This paper presents an effective method to estimate the permittivity profile from the scattered field measured outside scatterers by minimizing the non-radiating objective function using Monte Carlo approach. Since the non-radiating objective function is inherently non-linear as reported in [1], it is necessary to use some random perturbation to prevent the permittivity profile estimation from getting trapped in a local minimum. To do so, we have employed a Monte Carlo approach to minimize the non-radiating objective function by searching over the solution space. The results indicate the Monte Carlo approach can converge to the correct profile despite the existence of a few local minima within the solution space.

Scatterer localization based on the non-radiating equivalent source

This paper presents a numerical approach for a scatterer localization based on the non-radiating equivalent source. We have been able to localize a scatterer by minimizing the non-radiating objective function and applying some constraints. Simulation is also used to validate its performance. The scatterer localization approach can be utilized as an EMISS calibration process for high-frequency tomographic imaging.

Electromagnetic inverse scattering system characterization based on green's function analysis

This paper presents a very simple and effective numerical approach for an electromagnetic inverse scattering systems (EMISS) characterization based on 2D Greens function analysis. For this purpose, a number of Greens function matrices are constructed for various random distributions of source elements and observation point positions. We have employed the numerical rank of the Greens function matrices as the degree-of-uniqueness of the EMISS. Plotting the degree-of-uniqueness vs. the number of source EMISS elements enables us to compare different EMISS settings and find the optimum one for an EMISS implementation. The extensive simulation results confirm the effectiveness of the Greens function characterization curves.

Stand-alone embedded vision system based on fuzzy associative database

In this paper, a design methodology for a stand-alone embedded vision system (SEVS) is presented. The combination of region-based features and fuzzy theory defines the system, which is fast, flexible, and efficient. The proposed system can help to achieve flexible manufacturing goals and enhance safety. The advantages of the proposed system over traditional non-imaging sensors for manufacturing purposes include the recognition of the incoming product prior to determining its position, orientation, and speed. Region-based features - such as, Zernike moments, the first invariant function of central moments, and compactness - are utilized as pose descriptors. Moreover, we study the robustness of the pose descriptors and compare the fuzzy associative database (FAD) with maximum likelihood (ML) and a radial-basis function network to achieve multiple-pose detection. In addition, an ML estimation is employed to train the system automatically. It is demonstrated that the system can reliably recognize products with fairly complex shapes. When a product is successfully recognized, the system provides the essential information to a process controller or programmable logic controller for further action without requiring any particular interface. In the case of unrecognized objects, the system sends an appropriate message to the controller.