A Method of Aircraft Detection Using Fully Convolutional Network

Traditional methods on aircraft detection in remote sensing images rely on handcrafted design, which is difficult to detect and recognize the target in complex scenes and multiscale conditions. In this paper, we tackle these two problems by proposing a method for aircraft detection based on the fully convolutional neural network(FCNN). FCNN can obtain the location of the aircraft quickly and directly by minimizing a multi-task loss. Through data augmentation and transfer learning, the classification accuracy of FCN is much improved. In order to recognize small targets, we combine the resolution information with priori knowledge of aircraft to construct image pyramid structure on the test images. The experimental results show that the method with less parameters has higher accuracy and the model is simple to train.