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.

660 CHAPTER 14 Automatic Target RecognitionSome waveform decisions include whether or not to use stepped frequencies, pulse compression,stretch processing and bandwidth segmentation. Additionally, Tait presents agood summary <strong>of</strong> the effect <strong>of</strong> waveform choice on range window, time on target, rangeambiguities and operation in clutter environments.An additional consideration for feature observation is that performance for HRRPsis largely based on knowing the aspect angle <strong>of</strong> the target <strong>of</strong> interest, since scatters canchange significantly. To address this problem, normally a target is traced for a short timeto determine its direction before doing range-pr<strong>of</strong>ile measurements. With an approximatedirection, the number <strong>of</strong> aspect angles to be considered is reduced.An equally challenging issue is clutter suppression, which is <strong>of</strong>ten necessary to distinguishthe target. Kahler and Blasch [85] discuss several techniques for mitigating theeffects <strong>of</strong> clutter, including Doppler filtering, space-time adaptive processing (STAP)and displaced phase center arrays (DPCA). The Doppler filtering approach can be usedwith adaptive radars that are capable <strong>of</strong> detecting the distribution <strong>of</strong> clutter and adjustingthe radar appropriately to minimize the signal-to-clutter ratio. STAP uses Doppler frequency,platform velocity and direction <strong>of</strong> arrival information to reduce clutter. Displacedphase center arrays compensate for radar motion to reduce the Doppler spread <strong>of</strong> groundclutter.Additional difficulties include the existences <strong>of</strong> multiple or extended targets, whichcan go against the initial assumptions <strong>of</strong> the ATR system. Some HRR approaches aremore susceptible to problems related to moving targets, and thus require the data to bemotion compensated, or mo-comped (a process <strong>of</strong> measuring target velocity and adjustingthe phase accordingly). Additionally, data compression must occur for real systems[86]. Unfortunately, compression is expensive, and involves calibration and measurementerrors. Unpredictable experimental parameters and a lack <strong>of</strong> realistic target datafor all configurations and scenarios also add to the complexity <strong>of</strong> the data compressionproblem.14.7.4 Step 4: Test the Feature SetA number <strong>of</strong> techniques are given in the literature for testing HRRPs, even if only as asecondary feature in a broader ATR scheme. Mitchell and Westerkamp [82] explain thetraditional approach as a constrained quadratic classifier. To implement this approach,compute an ensemble mean and variance estimate for each range bin in the HRRP signatures.Then discriminate based on the observation and statistics (mean and variance).This can also be implemented with a mean-squared error to eliminate the variance terms.They also present a statistical feature-based classifier [82–83] method, which builds onan idea originally developed for air target identification and is implemented by lookingat probabilities calculated with a peak location probability function. Similarly, Jacobsand O’Sullivan [81, 76] use a likelihood approach, combining dynamics based on targetorientation and likelihoods <strong>of</strong> pre-determined pr<strong>of</strong>iles, given the orientation. Still others[84–89] choose a Mahalanobis distance metric that treats HRRR measurements as a vector<strong>of</strong> Gaussian elements.Layne [90] presents a multiple model estimator (MME) approach with multipleKalman Filters (KF). His work builds on previous work that considered MMEs for SARmeasurements [76, 91–92]. Layne’s approach is to use one KF for each target type and estimatekinematics (including position, velocity, and possibly acceleration). He then is ableto predict the target based on likelihoods from the various filters. The filters are designed to

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

Saved successfully!

Ooh no, something went wrong!