Abstract

One of the first steps within an automatic target recognition (ATR) system is to locate and then discriminate the desired moving target from from stationary clutter of the image. Due to the image formation processing, the detection of slow moving targets is a challenging problem with single phase center radar systems. If the Doppler shift due to target’s radial velocity is insufficient to displace it from the endo-clutter region then the signature must compete with generally brighter background rather than the power to noise floor in the exo-clutter region. Many commercial satellite vendors now provide long-dwell mode collections to create video SAR or colorized subaperture imagery (CSI) products. CSI is formed from the single look complex (SLC) image by applying a colormap across the aperture axis in the spatial frequency domain to create a three channel image. The advantage with this method compared to video SAR techniques is to preserve cross-range resolution for multilook speckle reduction. After CSI processing, clutter within the image is colored when the return of energy is not uniform across the aperture. The results of CSI processing of simulated scenes as well as open data from commercial collections will be presented. In this thesis, an automatic technique is detailed to detect suspected moving targets by exploiting colorized imagery with a segmentation technique based on a graph theoretic approach. The algorithm is designed to find gradients of colored pixels, as the azimuthal defocused signature of moving targets are colored to match the gradient of the chosen colormap. This moving target signature is verified by comparing CSI processed chips of a simulated point targets to the signature of ships in commercial imagery with a known speed and heading. The effect of this coloring will be shown to diminish with coarser resolution in sliding spotlight or stripmap modes in comparison to the long dwell spotlight imagery. A method was developed to filter the hue, saturation and value channels that removes approximately 99% of background pixels. A k nearest neighbors (k-NN) graph is constructed from the filtered image, followed by a graph based partitioning method with recursive bisection. The individual segments are evaluated and if suspected to be moving targets with the characteristic color gradient each would be passed onto the next steps of the ATR processing chain. The detection algorithm is shown to efficiently identify the color gradients within a variety of commercial imagery chips for varying signal to clutter ratios (SCR). The major limitation of this technique is the inability to detect targets with purely radial velocity, as the signature will not appear colored after processing.

Publication Date

7-31-2026

Document Type

Thesis

Student Type

Graduate

Degree Name

Imaging Science (MS)

Department, Program, or Center

Chester F. Carlson Center for Imaging Science

College

College of Science

Advisor

Charles Bachmann

Advisor/Committee Member

James Albano

Advisor/Committee Member

John Kerekes

Campus

RIT – Main Campus

Download

Share

COinS