Target Detection From Synthetic Aperture Radar (SAR) Image using Subspace Filtering:

We proposed a new class of Subspace filter based algorithms for detecting targets in forest clutter environment. The training phase of the proposed SAR target detection algorithms “learns” the clutter characteristics using local or global clutter subspaces. Both off-line and on-the-fly self-training versions of the algorithm are investigated. These adaptive approaches utilize the Singular Value Decomposition (SVD) algorithm where small blocks of data in the neighborhood of a sliding test window are processed in real-time to estimate clutter characteristics. The real-time clutter models are then used to nullify clutter in the test window. The proposed approach is shown to be highly effective for removing blob-like impulsive clutter for improved detection performance. An ultra-wideband (UWB) synthetic aperture radar (SAR) simulation technique that employs physical and statistical models was also generated. This joint physics-statistics based technique generates realistic images that have many of the “blob-like” and “spiky” clutter characteristics of UWB radar data in forested regions while avoiding the intensive computations required for the implementation of low-frequency numerical electromagnetic simulation techniques. The SAR image simulation technique allows system designers to test in a variety of operating environments. The proposed subspace filter-based detection algorithms have been applied on the simulated UWB data successfully.

Automatic Target Recognition:

This project addresses a new hybrid Automatic Target Recognition (ATR) algorithm on high range resolution (HRR) profiles. The proposed hybrid algorithm combines Eigen-Template based Matched Filtering (ETMF) and Hidden Markov modeling (HMM) techniques to achieve superior HRR-ATR performance. In the algorithm, each HRR test profile is first scored by ETMF, which is then followed, by independent HMM scoring. The first ETMF scoring step produces a limited number of "most likely" models that are target and aspect dependent. These reduced numbers of models are then used for improved HMM scoring in the second step. Finally, the individual scores of ETMF and HMM are combined using Maximal Ratio Combining to render a classification decision. The results demonstrate that the hybridization technique achieves improved recognition performance when compared to the independent performances of either ETMF or HMM. Performance comparison results are computed for both Forced decision and unknown target scenario case. The unknown target scenario is simulated using Leave One Out method (LOOM). The performance of ATR algorithms are compared in terms of Receiver Operating Characteristics (ROC) curves. The MSTAR data set is used for all simulations.

Target/Clutter Mathematical Modeling in Stochastic Scenario:

This project deals with Target and Clutter statistical modeling for Foliage Penetrating (FOPEN) Radar. A new bimodal technique is developed for modeling ultra wideband radar clutter. One type of model was developed for purposes of simulating data in the neighborhood of bright scatterers and another type of data model was developed for simulating the "residual" component of the clutter. These two models are combined to generate synthetic clutter images. Targets chips are synthesized by applying a radar scattering center model that is based on the geometrical theory of diffraction. Targets are modeled with either one or two dihedrals. A new adaptive rank-order filter algorithm is developed to estimate the high-end of the clutter distribution, or "clutter tail". This filtering approach is denoted as the "Discontinuity Filter". The results are shown in the form of Radar Operating Characteristics (ROC) plots. The simulation results indicate that, while the baseline Constant False Alarm Rate (CFAR) performance degrades significantly with increasing clutter density, the discontinuity filter shows some indication for maintaining a certain level of target detection performance.

Speaker Recognition:

The aim of this work is to build an Improved Speaker Recognition system by F0 Manipulation. There we built a Text Dependent Speaker Recognition System using a small speech database with varying vocal effort levels. We use Vector Quantization Distortion (VQD) and HMM for Speaker Recognition. In real-world conversations, talkers constantly change the levels of their voices either to add inflection to their speech or to be heard at a comfortable level by nearby or distant listeners. Our aim to make the automatic speech recognition software was to examine the effects due to change in vocal effort level on speech recognition. The model is trained with conversational speech and tested with shouted speech and vice versa. The aim is to improve the recognition rate using the manipulated speech files by including the pitch information of the files to be tested into the training phase.