遥感图像上的飞机目标检测是一件极富挑战性的工作，吸引了广大研究者的兴趣。基于人工神经网络的方法是当前遥感图像飞机目标检测的主流方法，这类方法要求人工标记大量的数据用于训练。对训练图像的人工标注工作费时费力，是制约有效利用大规模数据的主要瓶颈之一。为解决这个问题，提出一种基于生成对抗网络（generative adversarial networks，GAN）的半监督检测方法。在遥感图像飞机目标检测中，该方法不需要标记全部用于训练的图像，只需要标记其中一小部分样本，再和大量未标记数据一起进行训练便能取得优异的检测结果。该方法结合传统的检测网络和基于GAN的半监督学习网络。在对抗训练过程中，生成器学习数据分布并生成假样本，判别器判别真假样本，同时判别器还需要从标记数据中学习类别信息。最后，判别器学习到的决策分类面不仅仅区分出标记数据，而且平行于数据分布的边界。实验证明，在存在大量可供训练的图像的基础上，减少标注数据的比例，全监督学习方法性能会大幅下降；而本文提出的半监督学习方法，由于利用了未标注的数据，能保持更好的检测性能。
Airplane detection in remote sensing images is a challenging task and researchers have been sparing no efforts in making breakthroughs in this topic. The methods based on artificial neural network are the main methods for airplane detection in remote sensing images. However, a large number of data must be annotated for training, which is time-consuming. This is one of the main bottlenecks to limit large-scale data to be effectively used. To address this problem, a generative adversarial networks (GAN)-based semi-supervised object detection method is proposed in this work. In airplane detection in remote sensing images, the method does not need to annotate all of the images, but only a small part of them, and then trains them together with a large number of unannotated data to achieve excellent detection performance. The proposed method combines a traditional detection network with a semi-supervised learning network based on GAN. During the adversarial training procedure, the generator learns the data distribution and generates fake samples, while the discriminator distinguishes between the true samples and the generated fake samples. At the same time, the discriminator is also trained on labeled data. Finally, the discriminator learns the decision boundary that not only separates labeled samples, but also parallels with the distribution boundary of the whole data. Experiments show that, with a large number of images, reducing the proportion of labeled data will significantly reduce the performance of the full-supervised learning method, while the semi-supervised learning method proposed in this work maintains the detection performance due to the use of unlabeled data.
[1] Lowe D G. Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004, 60(2):91-110.
[2] Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]//International Conference on Computer Vision & Pattern Recognition (CVPR'05). IEEE Computer Society, 2005:886-893.
[3] Felzenszwalb P, Mcallester D, Ramanan D. A discriminatively trained, multiscale, deformable part model[C]//IEEE Conference on Computer Vision & Pattern Recognition, 2008:1-8
[4] Ren S, He K, Girshick R, et al. Faster R-CNN:Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015, 39(6):1137-1149.
[5] Redmon J, Divvala S, Girshick R, et al. You only look once:unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016:779-788.
[6] He K, Zhang X, Ren S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2014, 37(9):1904-1916.
[7] Liu W, Anguelov D, Erhan D, et al. Ssd:single shot multibox detector[C]//European Conference on Computer Vision. Springer, Cham, 2016:21-37.
[8] Liu L, Pan Z, Lei B. Learning a rotation invariant detector with rotatable bounding box[J]. arXiv preprint arXiv:1711.09405, 2017:1-9.
[9] Song Z N, Sui H G. a coarse-to-fine model for airplane detection from large remote sensing images using saliency modle and deep learning[J]. International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences, 2018, 42(3):1571-1573.
[10] Yu Y, Guan H, Zai D, et al. Rotation-and-scale-invariant airplane detection in high-resolution satellite images based on deep-Hough-forests[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2016, 112:50-64.
[11] Wu H, Zhang H, Zhang J, et al. Fast aircraft detection in satellite images based on convolutional neural networks[C]//2015 IEEE International Conference on Image Processing (ICIP). IEEE, 2015:4210-4214.
[12] Chen Z, Zhang T, Ouyang C. End-to-end airplane detection using transfer learning in remote sensing images[J]. Remote Sensing, 2018, 10(1):139-139.
[13] Yang Y, Zhuang Y, Bi F, et al. M-FCN:effective fully convolutional network-based airplane detection framework[J]. IEEE Geosci Remote Sens Lett, 2017, 14(8):1293-1297.
[14] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014:580-587.
[15] Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets[C]//Advances in Neural Information Processing Systems. 2014:2672-2680.
[16] Salimans T, Goodfellow I, Zaremba W, et al. Improved techniques for training gans[C]//Advances in Neural Information Processing Systems. 2016:2234-2242.
