Swapan Chakrabarti

Visualizing radiation patterns of antennas

An efficient computer graphics tool for investigating the radiation patterns of antennas as functions of their design parameters is presented. Radiation patterns (or the magnitudes of field distributions) of antennas are treated as three-dimensional surfaces, since the magnitudes of the far-field distributions are functions of polar and azimuthal angles. Two-dimensional GKS (Graphics Kernel Systems) routines and a shading scheme based on the full-wave solution of light scattering from rough surfaces are used to provide a realistic view on the computer screen. Different types of radiation patterns and their characteristics are discussed. A review of antenna fundamentals is included.<<ETX>>

Artificial neural network--based analysis of high-throughput screening data for improved prediction of active compounds.

Artificial neural networks (ANNs) are trained using high-throughput screening (HTS) data to recover active compounds from a large data set. Improved classification performance was obtained on combining predictions made by multiple ANNs. The HTS data, acquired from a methionine aminopeptidases inhibition study, consisted of a library of 43,347 compounds, and the ratio of active to nonactive compounds, R A/N, was 0.0321. Back-propagation ANNs were trained and validated using principal components derived from the physicochemical features of the compounds. On selecting the training parameters carefully, an ANN recovers one-third of all active compounds from the validation set with a 3-fold gain in R A/N value. Further gains in RA/N values were obtained upon combining the predictions made by a number of ANNs. The generalization property of the back-propagation ANNs was used to train those ANNs with the same training samples, after being initialized with different sets of random weights. As a result, only 10% of all available compounds were needed for training and validation, and the rest of the data set was screened with more than a 10-fold gain of the original RA/N value. Thus, ANNs trained with limited HTS data might become useful in recovering active compounds from large data sets.

Application of special-purpose artificial neural networks for speckle reduction in SAR images

Synthetic aperture radar (SAR) is used extensively for remote-sensing applications due to its ability to operate under all weather conditions and provide high-resolution images. However, high-resolution images constructed from SAR data often suffer from speckle, which makes identification and classification of edges/boundaries a difficult task. Speckle noise is multiplicative in nature and is a result of constructive and destructive interference of signals from randomly distributed scatterers in a resolution cell illuminated by a coherent signal. Usually, speckle is reduced by incoherent averaging of high-resolution image pixels that degrade resolution. The principal goal in all speckle-reduction algorithms is to reduce speckle with minimum loss of resolution. In this investigation, we used specially trained and validated artificial neural networks (ANNs) for speckle reduction in images generated with a radar-depth sounder/imager and compared their performance to the conventional adaptive filtering and Speckle Reducing Anisotropic Diffusion (SRAD) algorithm. We show that by training different ANNs to reduce speckle noise at different levels of signal-to-noise ratio (SNR), rather than training one ANN to operate at all levels of SNR, improved performance in speckle reduction can be obtained. Real SAR images and synthetic noise are used in this research to compare the performance of the proposed ANN-based approaches with that obtained from conventional methods. This investigation shows that on combining the results from a set of properly trained and validated neural networks, the SNRs of the output images improve beyond those obtained from conventional approaches when the input SNRs are greater than or equal to 4 dB. For input SNRs greater than 0 dB, however, the ANNs provide better performance in edge preservation compared with conventional methods. We also found that once a set of ANNs is properly trained to reduce speckle from an image, these ANNs can be used in de-speckling other images without any further training. The merits and demerits of different configurations of the ANNs are studied to find useful speckle noise-tolerant ANN architectures.

A fuzzy neural-network-model for aspect-independent target identification

A neural network is trained, using the fundamental properties of fuzzy-set theory, to achieve robust aspect-independent radar target identification. The radar cross section of two different aircraft are modeled using a thin-wire-time-domain (TWTD) code to compute their backscattered electric fields for twenty five different aspect angles. The scattered fields corresponding to a few aspect angles are then used to train the network and the rest of the scattered fields are used to test the performance of a neural network for target identification. A fuzzy neural network is found to provide superior performance for target identification compared with both a conventional neural network and a statistical Bayes classifier, especially in a noisy environment.<<ETX>>

Neural network‐based classification of scenes from SAR images using spectral information: An empirical study

The use of a feed‐forward artificial neural network (FANN) in classifying scenes from a SAR image, acquired over the KUREX test sites, is presented. Three different types of scenes (river, forest, and grassy fields) are located on the SAR image using an optical image and a ground map. For each type of scene, one hundred segments are located with each segment consisting of 16 x 16 pixels. The texture information of each segment of the image is obtained by computing the spectrum of its intensity distribution, after removing the mean intensity from the individual pixel intensities. A feature vector is then obtained for each segment using 64 samples of the spectrum and the mean value of the intensity distribution of the image segment. Ten different feature vectors from each type or class of scene are used to train a FANN, and the performance of the network is tested using the feature vectors that are not used during the training process. Different types of network architectures are considered in a search for ...

Neural network classification of subsurface targets

A classification technique for classifying subsurface targets based on the measured electric fields was developed in this paper. It was observed that subsurface targets can successfully be classified by using the magnitude of the scattered fields with the offset-VSP setup. The performance of the classification scheme was investigated with three different classification algorithms. It was observed that a neural network used with a distance criterion provided the best performance for this application. Use of the classical Bayesian classifier results in more misclassification than the neural network algorithms. The performance of both neural network algorithms exceeded the performance of the classical Bayesian classifier. The effect of the type of training data on the performance of the classifier was also studied in this paper. Training with a broad range of SNRs results in better classification rates as expected. But classifiers trained at 10 dB SNR also result in comparable performance. This unusual behavior shown by the network in this investigation is valid only for this specific application, and is probably due to the type of data used for classification.<<ETX>>