Satish S. Udpa

Texture classification using rotated wavelet filters

We propose an approach to the texture classification problem using a set of two-dimensional (2-D) wavelet filters that are nonseparable and oriented for improved characterization of diagonally oriented textures. Channel energies are estimated at the output of both the new filter bank and a standard discrete wavelet frames (DWF) filter bank. Classification results obtained using each individual method and in combination are presented. The results show that the oriented filter set results in finer discrimination providing complementary texture information to the DWF by making use of its orientation selectivity. As a result, a combination of the features from the output of two filter banks improved the classification accuracy significantly with a smaller number of features.

Characterization of gas pipeline inspection signals using wavelet basis function neural networks

Magnetic flux leakage techniques are used extensively to detect and characterize defects in natural gas transmission pipelines. This paper presents a novel approach for training a multiresolution, hierarchical wavelet basis function (WBF) neural network for the three-dimensional characterization of defects from magnetic flux leakage signals. Gaussian radial basis functions and Mexican hat wavelet frames are used as scaling functions and wavelets respectively. The centers of the basis functions are calculated using a dyadic expansion scheme and a k-means clustering algorithm. The results indicate that significant advantages over other neural network based defect characterization schemes could be obtained, in that the accuracy of the predicted defect profile can be controlled by the resolution of the network. The feasibility of employing a WBF neural network is demonstrated by predicting defect profiles from both simulation data and experimental magnetic flux leakage signals.

Ultrasound Image Texture Analysis for Characterizing Intramuscular Fat Content of Live Beef Cattle

The primary factors in determining beef quality grades are the amount and distribution of intramuscular fat percentage (IMFAT). Texture analysis was applied to ultrasound B-mode images from ribeye muscle of live beef cattle to predict its IMFAT. We used wavelet transform (WT) for multiresolutional texture analysis and second-order statistics using a gray-level co-occurrence matrix (GLCM) technique. Sets of WT-and GLCM-based texture features were calculated from ultrasonic images from 207 animals and linear regression methods were used for IMFAT prediction. WT-based features included energy ratios, central moments of wavelet-decomposed subimages and wavelet edge density. The regression model using WT features provided a root mean square error (RMSE) of 1.44 for prediction of IMFAT using validation images, while that of GLCM features provided an RMSE of 1.90. The prediction models using the WT features showed potential for objective quality evaluation in the live animals.

Invariance transformations for magnetic flux leakage signals

Magnetic flux leakage (MFL) methods are used extensively for inspecting ferromagnetic materials. The analysis of the MFL signal is however fraught with problems associated with the sensitivity of the signal to a number of factors such as the MFL sensor velocity and variations in the permeability of the test specimen. The interpretation can be simplified if the signals can be processed to attain invariance to these conditions. This paper presents novel methods for obtaining permeability invariant and velocity invariant MFL signals.

Solution of inverse problems in electromagnetics using Hopfield neural networks

Inverse problems are encountered in many fields of science and engineering. In electromagnetics, for example, inverse problems may involve the reconstruction of the source or scatterer on the basis of information contained in electromagnetic measurements. In general, the measurements can be related to the scatterer via integral equations. This paper presents a neural network approach for solving the inverse problem associated with such equations. Results of implementing the method on an application problem are presented. >

LEARN++: an incremental learning algorithm for multilayer perceptron networks

We introduce a supervised learning algorithm that gives neural network classification algorithms the capability of learning incrementally from new data without forgetting what has been learned in earlier training sessions. Schapires (1990) boosting algorithm, originally intended for improving the accuracy of weak learners, has been modified to be used in an incremental learning setting. The algorithm is based on generating a number of hypotheses using different distributions of the training data and combining these hypotheses using a weighted majority voting. This scheme allows the classifier previously trained with a training database, to learn from new data when the original data is no longer available, even when new classes are introduced. Initial results on incremental training of multilayer perceptron networks on synthetic as well as real-world data are presented in this paper.

Magnetic flux leakage modeling for mechanical damage in transmission pipelines

This paper presents a two stage FE model for prediction of magnetic flux leakage, resulting from mechanical damage. In the first stage the stress distribution associated with mechanical damage is obtained from a structural model. In the second stage the stress distribution is incorporated into a magnetic FE model, by mapping stress levels to permeability. MFL signals are calculated and compared with experimental gouge MFL signatures.

Solution of inverse problems in electromagnetic NDE using finite element methods

This paper presents a technique for solving inverse problems in magnetostatic nondestructive evaluation (NDE), using finite element models. In-line inspection of ferromagnetic gas pipelines containing pipe-wall defects, is chosen as the candidate NDE process. The signal inversion technique consists of iteratively solving the forward problem by updating the finite element mesh, rather than the material properties of the finite elements. Preliminary simulation results obtained using a 2D finite element model are presented.

Wavelet-based permeability compensation technique for characterizing magnetic flux leakage images

The magnetic flux leakage method, used for nondestructive evaluation of ferromagnetic objects, generates greyscale images that are representative of the integrity of the specimen. Defective areas typically appear as bright regions in the image. Unfortunately, the task of defect characterization becomes more challenging due to the effects of variations in the test parameters associated with the experiment. One such test parameter is the permeability of the test object. Conventional invariant pattern recognition algorithms are not capable of performing invariance transformations to compensate for such variations. This paper describes a novel technique that uses wavelet basis functions to provide selective invariant features and eliminate image intensity variations from undesirable changes in operational variables. The performance of the invariance transformation is demonstrated by applying the method to magnetic flux leakage images obtained using a finite element simulation of in-line inspection of natural gas transmission pipelines.

3D simulation of velocity induced fields for nondestructive evaluation application

The challenges associated with finite element modeling of a tight crack in an otherwise large geometry encountered in a typical nondestructive evaluation application are considered. A novel approach for decomposing the modeling of velocity induced fields in a conducting ferro- or nonferromagnetic material is presented.