Airplane detection using convolutional neural networks in a coarse-to-fine manner
Xiaobin Li, Shengjin Wang, Bitao Jiang, Xiaobing Chan · 2017 IEEE 2nd Information Technology, Networking, Electronic and Automation Control Conference (ITNEC) · 2017
Airplane detection in remote sensing images is a challenging task due to the diversity of airplanes and the complexity of backgrounds. In this paper, we propose an airplane detection method using convolutional neural networks (CNNs) in a coarse-to-fine manner which simulates the detection manner of image analysts. Our method proposes coarse candidate regions containing multiple airplanes first, then finely detects each airplane in these candidate regions. According to this manner, we design a precise and efficient detection framework which consists of two CNNs with the same structure. One CNN is used to coarsely propose candidate regions, the other is used to finely detect airplanes. Using this method, we can generate fewer candidate regions than the existing literatures and extract discriminative deep features. Experiments on Google Earth images demonstrate that our method is accurate and efficient.