Rajesh Shenoy

Feature space trajectory for distorted-object classification and pose estimation in synthetic aperture radar

Classification and pose estimation of distorted input objects are considered. The feature space trajectory representation of distorted views of an object is used with a new eigenfeature space. For a distorted input object, the closest trajectory denotes the class of the input and the closest line segment on it denotes its pose. If an input point is too far from a trajectory, it is rejected as clutter. New methods for selecting Fukunaga-Koontz discriminant vectors, the number of dominant eigenvectors per class and for determining training, and test set compatibility are presented.

Synthetic aperture radar detection and clutter rejection minace filters

Abstract New shift-invariant distortion-invariant Synthetic Aperture Radar (SAR) filters for detection are described. They employ new modified Minimum Noise and Correlation Energy (MINACE) filters. Test results are provided for real SAR data with 360° aspect views of two different objects and for 100 real SAR clutter chips (natural and man-made) that passed the second stage of the Lincoln Laboratorys SAR processor. We obtained high detection rates P D =100% and few false alarms P FA ⩽ 0.82% using only three filters per object class. The shift-invariance of the filters allow their use as adjuncts for all stages (detection, discrimination, classification) of a SAR image processor. Classification results are also briefly reported.

Eigen-MINACE SAR detection filters with improved capacity

Distortion-invariant correlation filters are used to detect and recognition distorted objects in scenes. They are used in a correlator and are thus shift-invariant. We describe a new way to design distortion-invariant correlation filters that ensures good generalization (same performance on training and test sets) and improved capacity (fewer filters that recognize distorted versions of multiple classes of objects). The traditional way of designing correlation filters uses different types of frequency domain preprocessing and linear combination of training images. We show that these different approaches can be implemented in a framework using linear combination of eigen-images of preprocessed training data. Using eigen-domain data is shown to produce filters that generalize better and have large capacity. We show results on SAR data with multiple classes of objects using eigen-MINACE filters.

Feature space trajectory (FST) neural network for SAR detection, classification, and clutter rejection

We consider use of eigenvector feature inputs to our feature space trajectory (FST) neural net classifier for SAR data with 3D aspect distortions. We consider its use for classification and pose estimation and rejection of clutter. Prior and new MINACE distortion-invariant and shift- invariant filter work to locate the position of objects in regions of interest is reviewed. Test results on a number of SAR databases are included to show the robustness of the algorithm. New results include techniques to determine: the number of eigenvectors per class to retain, the number and order of final features to use, if the training set size is adequate, and if the training and test sets are compatible.

Multiclass SAR feature space trajectory (FST) neural network class and pose estimation results

The feature space trajectory representation and neural network is used for classification and pose estimation of distorted objects in SAR data. New feature spaces and techniques to extend the concept to multiple classes are emphasized with initial four class results. On 4 class data, we obtain Pc equals 98.3 percent and clutter PFA equals 0.026/km2.

Correlation filters that generalize well

Distortion-invariant correlation filters are used to detect and recognize distorted objects in image scenes. We describe a new way to design distortion-invariant correlation filters that ensures good generalization (same performance on training and test sets). The traditional way of designing correlation filters uses different types of frequency domain preprocessing and linear combination of training images. We show that these different approaches can be implemented in a framework using linear combination of eigen-images of preprocessed training data. Using eigen-domain data is shown to generalize well regardless of preprocessing used. We show results on SAR data using eigen-MINACE filters.

Eigen detection filters that reject unseen clutter

Detection requires locating all objects in a scene independent of their class or aspect view and rejecting clutter. We consider new eigen filters to achieve this. They can be implemented on optical or digital correlators. Shift-invariance is required. These filters must also reject unseen clutter. Test results are presented for multi-class Synthetic Aperture Radar (SAR) data.

Fast nondyadic shift-invariant Gabor wavelets

We consider several new applications of wavelet transforms: general time-frequency analysis, detection and range-Doppler processing. These differ from the standard data compression and encoding applications generally considered. We note problems with standard and dyadic wavelet transforms concerning shift-invariance, frequency resolution, detection in noise at high frequencies where the bandwidth of the wavelet subband is large, and in range-Doppler processing. We consider Gabor wavelets and advance a new shift-invariant continuous wavelet filter with only O(N) processing required; we also advance new non-dyadic wavelet filters and an optical implementation for the general use (especially for range-Doppler processing).

Detection and classification in SAR using MINACE correlation filters

New shift-invariant distortion-invariant SAR filters for detection and classification are described. They employ new modified MINACE filters. Test results are provided for real SAR data with 360 degree aspect views of 2 different objects and for 100 real SAR clutter chips (natural and man-made) that passed the second stage of the Lincoln Laboratorys SAR processor. We obtained PD equals 100% and PFA <EQ 0.82%. The shift-invariance of the filters allow their use as adjuncts for all stages (detection, discrimination, recognition) of a SAR image processor.

Detection filters for visible high-resolution imagery

We consider new distortion-invariant filters (DIFs) to detect objects in high-resolution Electro-Optical (EO) visible imagery. EO data is a difficult detection problem, because only primitive features such as edges and corners are useful. No hot spots (present in IR data) or bright reflectors (present in SAR data) exist in EO data. We thus expect many false alarms when we try to detect objects in EO data. We use new eigen-detection filters because they are shift-invariant, require only few filters and can handle multiple target classes. Initial results show that our filters, when using zero-mean data, perform well on EO data.