A Research On Airplane Automatic Recognition For Optical Remote Sensing Images

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, we study the airplane recognition for optical remote sensing images including these methods of preprocessing to remote sensing images, segmenting the regions of airplane targets from remote sensing images, establishing the image-database and the characteristic-database for airplane target, recognizing the type of airplanes, etc.Firstly, we study the methods of preprocessing to remote sensing images and segmenting the regions of airplane targets from them. We use the linearity-filter and the median-filter, and then segment the airplane with the optimal threshold method.Secondly, with regard to the study of characteristic extraction, we establish a swatch-database including one training swatch-database and one testing swatch-database with ten types of airplanes' planforms under the condition of different scales, different view angles and simple affine transformations etc. And then, we extract the Hu invariant moment, the affine invariant moment, the ratio of long to broad and the compact degree from the images of the training swatch-database as the characteristics of airplane targets and analyze the performance of the description of these characteristics.Finally, we study the methods of recognizing the type of airplanes by the experiment of classifying these images from each of the three types swatch-database, the five types swatch-database and the ten types swatch-database of airplane targets which are selected from the testing swatch-database with the least-distance classifier, the neural network classifier and the tree classifier. As a result, the tree classifier is better than the other two.