Aircraft Type Recognition in Multi-view Optical Images Using Coarse-to-Fine Framework

Zhixiang Liang, Rui Yu, Yanshan Li

Abstract: Aircraft type recognition in multi-view optical images is a challenging task. The imaging perspective is vulnerable to aircraft posture, flying height, and other factors. Such issues often cause the large intra-class difference within the captured scenes. In this work, we propose a novel coarse-to-fine aircraft type recognition framework in multi-view optical images, which consists of two main parts: instance segmentation model and contour template matching model. In the coarse stage, a contour-based instance segmentation model is applied to extract the aircraft contour in coordinate pointset form. Then inner-distance shape context (IDSC) is adopted as contour feature descriptor for contour template matching. In the fine stage, we design a novel matching evaluation criterion to select the optimal matching template. Experimental results show that the proposed approach achieves the top1 accuracy of 71.3% and top3 of 90.8% on the established aircraft type testing set.