SAR - Recognition of military obcjects from radar pictures

Authors: inż. Wojciech Foksowicz, Przemysław Drosik

Introduction:

The aim of this project is to write a program that recognizes and classifies military objects. The input data are the radar pictures of these objects. At black and white bitmaps (radar pictures) a military object is represented by a set of bright points (pixels).

Picture1. Radar picture of a military truck.

Recognition and classification of a military object is done comparing an input model with the models from database.

Acquisition of knowledge:

Knowledge that is acquired from radar pictures must be the most characteristic and important data taken from them. According to the fact that a military object is represented by a set of bright points, they create a pattern of a model. There are also bright points (pixels) visible that represent shadow of an object and any other disruptions. Due to this and the fact that points representing an object brighter, fifty the most brigtenest points are taken. A simple function finds them returning coordinates and the values of RGB components of each of them. These three values for each of these 50 points are acquired for further processing.

Data transformation and analysis:

The acquired knowlegde must be transformed so that further calculations can be done. For each bitmap (military object) a record is created that consists of three fields: RGBvalue, positionx and positiony. The RGBvalue corresponds to the value of each RGB component (the bitmap is black and white so the components values are equial) while the positionx and the positiony are x and y coordinates respectively. On these data a ranking is made. First of all the points are sorted and secondly a rank of each point is calculated.

Sorting is done by the decreasing value of RGB components. The rank is done by a simple function which to the 1st brightenest point assigns the biggest rang (equal to 1) and to the 50th point the smallest rang (equal to 50). In a case when two or more points have the same RGB value their rangs are the same. Having these data tarnsformed (sorted and ranked) a final representation of brightenest points of a military object is ready. It now can undergo further calculations which are the recognition and the classification of an object.

Pattern recognition:

The recognition of an object is done by comaprison of an input military object with the models from the database. The comparison consists of two phases:

• calculating the intervals between each pair of points
• comparing the ranks of the RGB values of closest points

Interval is calculated from the coordinates using Euclides formula:

In result a matrix of intervals between each point is created. Rows of this matrix corespond to the sorted and ranked points of the model while colums correspond to the sorted points of the object. The interval between points is calculated in two ways: between values of coordinates and between ranks of coordinates.

From this matrix the smallest intervals between each pair of points are found. For each point of an object a closest point (meaning Euclides interval) of a model is found. Assigning to each point of the object the closest point of the model a new ranking is created. It is now not sorted by the decreasing RGB value but by the position of closest points in the ranking of object.

From both object and new model rankings the ranks of RGB values are extracted, creating two single rankings of RGB values. These two rankings are compared using the Kendall’s measure which calculates the intervals between those ranking. A Kendall’s coefficient is calculated from the formula:

where d_k(R,R*) is Kendall’s interval between matrix R and R* and is calculated from formula:

The result of this comparison is the value of tÎá-1, 1ñ. If t is equal to 1 it means that the two rankings are the same, if t=-1 it means that the rankings are completely different.