This thesis reports on research into the concept of Micro-Doppler Signature (-DS) based radar Automatic
Target Recognition (ATR) with a lot more contributions to general radar ATR methodology. The -DS based area of the research plays a role in three distinct areas: time domain classification frequency domain classification and multiperspective -DS classification which includes the introduction of a theory for that multistatic -DS. The contribution to general radar ATR may be the proposal of the methodology to permit better look at potential approaches and also to allow comparison between different studies.
The suggested methodology relies around a “black box” type of a radar ATR system that, critically, features a threshold to identify inputs which are formerly unknown somewhere. Out of this model some five evaluation metrics are defined. The metrics boost the knowledge of the classifier’s performance in the common possibility of correct classification, that reports how frequently the classifier properly identifies a port, to focusing on how reliable it’s, how capable it’s of generalizing from
the reference data, and just how effective its unknown input recognition is. Furthermore, the value of performance conjecture is discussed along with a preliminary approach to estimate how good a classifier should perform is developed. The suggested methodology will be accustomed to assess the -DS based radar ATR approaches considered.
Time domain classification analysis relies around using Dynamic Time Warping (DTW) to recognize radar targets according to their -DS. DTW is really a speech processing technique that classifies data series by evaluating all of them with a pre-classified reference dataset.
This resembles the most popular k-Nearest Neighbour (k-NN) formula, so k-NN can be used like a benchmark by which to judge DTW’s
performance. The DTW approach is noted to be effective. It achieved high possibility of correct classification and reliability in addition to being in a position to identify inputs of unknown class. However, the classifier’s capability to generalize in the reference information is less impressive also it performed only slightly much better than an arbitrary selection in the possible output classes. Difficulties in classifying the -DS within the time domain are identified in the k-NN results prompting a big change towards the frequency domain.
Processing the -DS within the frequency domain allowed the introduction of a sophisticated feature extraction routine to maximise the separation from the target classes and for that reason lessen the effort needed to classify them. The regularity domain also allowed using the performance conjecture method developed included in the radar ATR methodology and the development of a nave Bayesian method of classification. The outcomes for that DTW and k-NN classifiers within the frequency domain were comparable
towards the time domain, an unpredicted result because it was anticipated the -DS could be simpler to classify within the frequency domain. However, the nave Bayesian classifier created excellent results that matched using the predicted performance suggesting it couldn’t be bettered. Having a effective classifier, that might be appropriate legitimate-world use, developed attention switched towards the options provided by the multistatic -DS.
Multiperspective radar ATR uses data collected from various target aspects concurrently to enhance
It’s been shown effective for a few of the options to -DS based ATR also it was therefore speculated it might enhance the performance of -DS ATR solutions.
The multiple perspectives needed for that classifier were collected utilizing a multistatic radar developed at College College London (UCL). Producing a dataset, and it is subsequent analysis, led to the very first reported findings within the novel field from the multistatic -DS theory. Regrettably, the character from the radar used led to limited micro-Doppler being noticed in the collected data which reduced its value for classification testing. An effort to make use of DTW to do multiperspective -DS ATR was
made however the outcome was inconclusive. However, thought on the enhancements provided by multiperspective processing in alternative types of ATR mean it’s still expected that -DS based ATR would
take advantage of this processing.
Radar target micro-doppler signature classification
Open access status:
A wide open access version can be obtained from UCL Discovery
Target recognition, radar, doppler, classification
Downloads since deposit
Archive Staff Only
College College based in london. Gower Street. London. WC1E 6BT Tel. +44 () 20 7679 2000