Classification of aircraft with shadow using keypoint descriptor from remote sensing images

Abstract

Presently, aircraft classification from remote sensing images is widely used in military and civilian. However, aircraft type classification is still challenging to discriminate. The reason is that the aircraft displayed in the image has different in color, shape, size, and orientation. Moreover, there is the shadow appeared over and behind the aircraft that obscures some details of aircraft. Therefore, this work proposes the classification of aircraft with shadow method. This method modifies the deeply supervised salient object detection with short connections to make the aircraft be outstanding from the shadow and other objects in the background and then segment it. The segmentation experiment performs on our dataset, which is divided into two categories: simple case and difficult case. The results show that our proposed method performs well in both cases. In order to classify the types of aircraft, we extract features from each keypoint of aircraft segmented image by using the Fourier descriptor based on centroid contour distance, which is invariant to position, scale, and rotation. After that, we use the random forest algorithm to classify the aircraft types into six classes in the classification experiment. The result shows that the method can efficiently classify the aircraft and work well on aircraft images with shadow.