Overview of Research

Current Research @ IIITS

PAST WORK

Postdoctoral Work @ UMES

1. Cognitive Radar: Parameter Estimation

At the University of Maryland Eastern Shore, I participated in a DoD and NSF funded research on the adaptive radar. My responsibilities included the development and implementation of adaptive radar signal processing algorithms. In this research, we consider the problem of sequential estimation of properties of an extended radar target with multi-antenna arrays using adaptive waveforms. The estimation problem is studied in a Bayesian framework to formulate sensing and estimation as an adaptive process based on prior knowledge of target models. Using iterative transmission of adaptive radar waveforms, the radar estimates the target parameters and updates its posterior probability (or beliefs) of the target model based on new measurements.

2. Free Energy Principle for Radar Signal Processing:

We proposed a unified approach for adaptive radar waveform design and target parameter estimation via the free-energy principle. The free-energy principle stems from the variational Bayesian approximation method in machine learning which aims to find an approximate distribution close to the true target distribution. In this research, this variational approach is applied to the estimation of a radar target’s mean response. The problem of parameter estimation and the waveform design are formulated as optimization problems under a common variational free energy objective function with different forms, which result in respective analytical solutions. We demonstrate that using adaptive waveforms, the convergence of the sequential estimation is accelerated. Furthermore, we show that for a univariate parameter estimation problem, the derived analytical solution is the same as the Bayesian posterior density. Moreover, the optimal waveform derived by the variational free energy approach is similar to the waveform derived by the well known mutual information criterion.

Postdoctoral Work @ MSU 

1. High resolution satellite image processing (2011-2012)

A hybrid approach is proposed for efficient building extraction from optical multi-angular imagery, where a template matching algorithm is formulated for automatic estimation of relative building height, and the relative height estimates are utilized in conjunction with a support vector machine (SVM)-based classifier for extraction of buildings from non-buildings. This approach is tested on ortho-rectified Level-2a multi-angular images of Rio de Janerio from WorldView-2 sensor. Its performance is validated using a 3-fold cross validation strategy. The final results are presented as a building map and an approximate 3D model of buildings. The building detection accuracy of the proposed method is improved to 88%, compared to 83% without using multi-angular information.

2. Detection of radio-active metals buried under ground using DSP on Electromagnetic Induction Data (2011)

Analysis of the data obtained from electromagnetic induction (EMI) sensors is one of the most viable tools for detection of metallic objects buried under soil. The early detection methods usually consisted of sophisticated EM modeling of the source/target geometry for building suitable discriminators. The major technical challenge in this field is reduction of false alarms with increase of detection probability. In this paper, we propose an unsupervised decision tree based one-class support vector machine (SVM) algorithm to detect buried radioactive targets, without sophisticated EM modeling. Using the EMI data obtained by a GEM-3 sensor, our proposed algorithm can more successfully detect the targets from non-target metals, compared to other unsupervised and supervised approaches.

PHD Work @ MSU

Main focus: Global Water cycle is one of the chief energy cycles that run the earth system. My PHD research is related to analysis and enhancement of datasets of hydrological parameters such as soil moisture and precipitation. Soil moisture is one of the most important environmental variables in regional weather and global climate systems. Precipitation, is also an important component of the global energy and water cycle; it is one of the main variables predicted in weather forecast models.  In this research, we propose spatio-temporal analysis methods to accomplish the following tasks: (i) consistency analysis of satellite-based soil moisture data, (ii) interpolation of missing data in soil moisture datasets, and (iii) merging of satellite-based precipitation observations. Novel pattern recognition approaches are developed in the first and  the third tasks. Existing signal processing methodologies are used and modified in the second task.