| Περίληψη: | Nowadays, a great number of researchers are concerned about aerial targets since they are involved in many aspects of everyday life, such as military issues, including aircrafts and missiles, or Unmanned Aerial Vehicles (UAVs) flying over city centers or airport areas. In this thesis, the design and the evaluation of an Automatic Target Recognition (ATR) system which efficiently derives and uses the bistatic radar crosssection (BS-RCS) information is presented. More specifically, the BS-RCS values are extracted using the Boundary Element Method (BEM) implemented in PITHIA software, a simulator that gives us the ability to efficiently compute the RCS at any point of the coordinate system. The discretization of the object surface is carried out also with the PITHIA software, which gives the nodes and elements that are necessary for the BEM. A sphere and spheroid are selected as object targets because of their axisymmetric geometry, which significantly reduces the computational cost and the time necessary for the computation of the BS-RCS value; a perfectly conducting sphere with radius r=0.76 m and a perfectly conducting spheroid with a=0.5 m and b=1.75 m. Afterwards, another spheroid with a=0.25 and b=0.72 is added which makes the classification process a much more difficult task. In our implementation any arbitrary target geometry can be used, such as UAVs, missiles or aircrafts. The angle between the ground and the propagation of electromagnetic plane wave is 2.86 degrees. For each receiver we compute the RCS, which is used for the classification of the targets. Furthermore, we consider 21 different points for each moving target in order to have a satisfactory number of RCS data. All the objects studied in this work are perfectly conducting and thus the radial scattering amplitude is zero. For each of the targets and for each point the RCS is computed for the frequencies: 90 MHz, 250 MHz, 800 MHz, 1.3 GHz. In the classification process, a modified k-means, in combination with k-NN algorithm and an artificial Neural Network (ANN) process both noise-free data and data whose RCS values are corrupted by noise.
|
|---|
