Aircraft Type Recognition in Remote Sensing Images Using Deep CNN

Aircraft type recognition is a meaningful task in remote sensing images. It remains challenging due to the difficulty of obtaining appropriate representation of aircrafts for recognition. To tackle these problems, a novel and robust aircraft type recognition framework based on Convolutional Neural Networks (CNN)s has been proposed in this paper. First, the proposed system focus on obtaining more detailed aircraft segmentation results without using refined annotations in the training stage. A convolutional encoder decoder network is designed to capture coarse segmentations. Then, a conditional random field (CRF) is used to refine the segmentation results. Second, to acquire more accurate direction estimations, the direction estimation is transformed to a keypoints’ detection task, and a convolutional regression network is built to locate the positions of aircraft’s keypoints. Besides, a multirotation refinement (MRR) method is proposed to further improve the precision of keypoints’ positions. At last, the template matching procedure is carefully designed to recognize aircrafts based on the direction estimations and segmentation results. The similarity between segmentation results and templates is evaluated by adopting the Intersection Over Union (IOU) measure. The proposed framework takes advantage of both shape and scale information of aircrafts for recognition. Experimental results show that the proposed method outperforms the state-of-the-art methods and can achieve 95.6% accuracy on the challenging data set.