Aircraft Recognition Algorithm Based On Rough Sets And Support Vector Machines

Automatic Target Recognition (ATR) is one of the important parts of computer vision, and has been widely applied for military. The airplane recognition is an application of ATR, and is a key technique in intelligence. Till now, there is no well-built theory for airplane recognition and it is still a challenge for the researchers.In this paper, technique of pattern recognition based on rough set and support vector machine, combined with the airplane recognition for optical remote sensing images is analyzed and researched. A classifier based on the advantage of rough set and DAGSVM (Directed Acyclic Graph Support Vector Machines) is designed. By means of comparable research and result analysis of classifier, some satisfied research fruit is obtained. The main content and results are shown as follow:Firstly, we establish a swatch-database including one training swatch-database and one testing swatch-database with planforms under the condition of different scales, different view angles of nine airplanes in different forms and three airplanes in similar forms by using 3DSMAX.Secondly, the value of invariant moment has many unstable factors when the condition of imaging is unchanged, especially, there will be a great error when computing the invariant moment for small dimension target. Accordingly, this paper studies the characteristic extraction method of Hu invariant moment of airplane gray image as well the complex number moment and establishes the multi-level invariant moment swatch-database based on dimension and blur factor. Experiment shows that the multi-level swatch-database provides a great improvement in the precision of ATR.Thirdly, the multi-class classification method of support vector machine is researched and their principles and characteristics as well inner relationships are analyzed in this paper. As the 1-v-1 and 1-v-r method does not present the analysis of error generalization ability and the relationship between the training time and node is non-linear, a DAGSVM classification arithmetic is used in this paper. It is proved in experiment that the classification precision is improved and the time of training and forecast is shorter.Fourthly, feature extraction based on rough set was researched. How the dimension of samples affected classification performances is analyzed. Accordingly,...