11.07.2015 Views

Principles of Modern Radar - Volume 2 1891121537

Principles of Modern Radar - Volume 2 1891121537

Principles of Modern Radar - Volume 2 1891121537

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

14.2 Unified Framework for ATR 633PDPEMSROCSARSVDSVMUHFprobability <strong>of</strong> detectionpredict, extract, match, and searchreceiver operating chacteristicsynthetic aperture radarsingular value decompositionsupport vector machineultra high frequencies14.2 UNIFIED FRAMEWORK FOR ATREven though the applications <strong>of</strong> ATR are diverse, the basic framework for ATR is remarkablysimilar and can be decomposed into four main steps, referred to here as the unifiedframework for ATR. The following sections describe these steps, as well as commonchallenges.14.2.1 Step 1: Identify the Target SetThe first step in development <strong>of</strong> an ATR algorithm is to identify the target set, whichincludes the targets <strong>of</strong> interest, as well as targets that are not <strong>of</strong> interest but are likely tobe observed. Implicit in this decision is the desired level <strong>of</strong> precision. For example, thegoal <strong>of</strong> some ATR systems is to distinguish between broad classes <strong>of</strong> targets, while thegoal <strong>of</strong> others is to distinguish specific models within a class. In the former case, the targetset may include broad classes such as aircraft, satellites, missiles, and unknown targets,whereas in the latter case it might include more precise classes <strong>of</strong> aircraft models such asF-15, F-16, T-38, and unknown targets. Subsequent decisions are heavily influenced bythis first step.14.2.2 Step 2: Select the Feature SetHaving identified the target set, the next step is to select a feature set that maximizes “thesimilarity <strong>of</strong> objects in the same class while maximizing the dissimilarity <strong>of</strong> objects indifferent classes” [118, pp. 367]. Sensitivity is a key component in this definition. Somefeatures (e.g., radar cross section, when the wavelength is much smaller than the target)are extremely sensitive to small changes in target parameters like aspect or pose; hence,they do not maximize the similarity <strong>of</strong> objects in the same class, or even <strong>of</strong> the same objectover time. The goal is to find a set <strong>of</strong> features sufficiently sensitive to reliably separatetargets into classes but not so sensitive that targets in the same class are appropriated intodisparate classes.14.2.3 Step 3: Observe the Feature SetOnce the feature set has been selected, the next challenge is to observe it as accuratelyas possible. Depending on the application and the choice <strong>of</strong> features, this may imply thatspecialized signal processing or conditioning (e.g., background removal) is necessary toextract observations <strong>of</strong> the feature set from the collected data. Accurately observing thefeature set may also require the sensor to operate in a non-standard mode (i.e., to usedifferent waveforms than are normally applied), which could drive fielding decisions.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!