skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Scalable Track Detection in SAR CCD Images

Abstract

Existing methods to detect vehicle tracks in coherent change detection images, a product of combining two synthetic aperture radar images ta ken at different times of the same scene, rely on simple, fast models to label track pixels. These models, however, are often too simple to capture natural track features such as continuity and parallelism. We present a simple convolutional network architecture consisting of a series of 3-by-3 convolutions to detect tracks. The network is trained end-to-end to learn natural track features entirely from data. The network is computationally efficient and improves the F-score on a standard dataset to 0.988, up fr om 0.907 obtained by the current state-of-the-art method.

Authors:
 [1];  [1]
  1. Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
Publication Date:
Research Org.:
Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
Sponsoring Org.:
USDOE National Nuclear Security Administration (NNSA), Office of Defense Nuclear Nonproliferation (NA-20)
OSTI Identifier:
1347496
Report Number(s):
SAND2017-2423
651687
DOE Contract Number:
AC04-94AL85000
Resource Type:
Technical Report
Country of Publication:
United States
Language:
English
Subject:
47 OTHER INSTRUMENTATION

Citation Formats

Chow, James G, and Quach, Tu-Thach. Scalable Track Detection in SAR CCD Images. United States: N. p., 2017. Web. doi:10.2172/1347496.
Chow, James G, & Quach, Tu-Thach. Scalable Track Detection in SAR CCD Images. United States. doi:10.2172/1347496.
Chow, James G, and Quach, Tu-Thach. Wed . "Scalable Track Detection in SAR CCD Images". United States. doi:10.2172/1347496. https://www.osti.gov/servlets/purl/1347496.
@article{osti_1347496,
title = {Scalable Track Detection in SAR CCD Images},
author = {Chow, James G and Quach, Tu-Thach},
abstractNote = {Existing methods to detect vehicle tracks in coherent change detection images, a product of combining two synthetic aperture radar images ta ken at different times of the same scene, rely on simple, fast models to label track pixels. These models, however, are often too simple to capture natural track features such as continuity and parallelism. We present a simple convolutional network architecture consisting of a series of 3-by-3 convolutions to detect tracks. The network is trained end-to-end to learn natural track features entirely from data. The network is computationally efficient and improves the F-score on a standard dataset to 0.988, up fr om 0.907 obtained by the current state-of-the-art method.},
doi = {10.2172/1347496},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Wed Mar 01 00:00:00 EST 2017},
month = {Wed Mar 01 00:00:00 EST 2017}
}

Technical Report:

Save / Share:
  • A Synthetic Aperture Radar (SAR) which employs direct IF sampling can significantly reduce the complexity of the analog electronics prior to the analog-to-digital converter (ADC). For relatively high frequency IF bands, a wide-bandwidth track-and-hold amplifier (THA) is required prior to the ADC. The THA functions primarily as a means of converting, through bandpass sampling, the IF signal to a baseband signal which can be sampled by the ADC. For a wide-band, high dynamic-range receiver system, such as a SAR receiver, stringent performance requirements are placed on the THA. We first measure the THA parameters such as gain, gain compression, third-ordermore » intercept (TOI), signal-to-noise ratio (SNR), spurious-free dynamic-range (SFDR), noise figure (NF), and phase noise. The results are then analyzed in terms of their respective impact on the overall performance of the SAR. The specific THA under consideration is the Rockwell Scientific RTH010.« less
  • SAR imagery for coastline detection has many potential advantages over conventional optical stereoscopic techniques. For example, SAR does not have restrictions on being collected during daylight or when there is no cloud cover. In addition, the techniques for coastline detection witth SAR images can be automated. In this paper, we present the algorithmic development of an automatic coastline detector for use with SAR imagery. Three main algorithms comprise the automatic coastline detection algorithm, The first algorithm considers the image pre-processing steps that must occur on the original image in order to accentuate the land/water boundary. The second algorithm automatically followsmore » along the accentuated land/water boundary and produces a single-pixel-wide coastline. The third algorithm identifies islands and marks them. This report describes in detail the development of these three algorithms. Examples of imagery are used throughout the paper to illustrate the various steps in algorithms. Actual code is included in appendices. The algorithms presented are preliminary versions that can be applied to automatic coastline detection in SAR imagery. There are many variations and additions to the algorithms that can be made to improve robustness and automation, as required by a particular application.« less
  • SAR range-Doppler images are inherently 2-dimensional. Targets with a height offset lay over onto offset range and azimuth locations. Just which image locations are laid upon depends on the imaging geometry, including depression angle, squint angle, and target bearing. This is the well known layover phenomenon. Images formed with different aperture geometries will exhibit different layover characteristics. These differences can be exploited to ascertain target height information, in a stereoscopic manner. Depending on the imaging geometries, height accuracy can be on the order of horizontal position accuracies, thereby rivaling the best IFSAR capabilities in fine resolution SAR images. All thatmore » is required for this to work are two distinct passes with suitably different geometries from any plain old SAR.« less
  • The objective of this work was to develop a systematic method of combining multifrequency polarized SAR images. It is shown that the traditional methods of correlation, hard targets, and template matching fail to produce acceptable results. Hence, a new algorithm was developed and tested. The new approach combines the three traditional methods and an interpolation method. An example is shown that demonstrates the new algorithms performance. The results are summarized suggestions for future research are presented.
  • SAR phase history data represents a polar array in the Fourier space of a scene being imaged. Polar Format processing is about reformatting the collected SAR data to a Cartesian data location array for efficient processing and image formation. In a real-time system, this reformatting or ''re-gridding'' operation is the most processing intensive, consuming the majority of the processing time; it also is a source of error in the final image. Therefore, any effort to reduce processing time while not degrading image quality is valued. What is proposed in this document is a new way of implementing real-time polar-format processingmore » through a variation on the traditional interpolation/2-D Fast Fourier Transform (FFT) algorithm. The proposed change is based upon the frequency scaling property of the Fourier Transform, which allows a post azimuth FFT interpolation. A post azimuth processing interpolation provides overall benefits to image quality and potentially more efficient implementation of the polar format image formation process.« less