针对建立和更新田间图形数据库时，以高分辨率遥感影像为底图人工勾绘地块耗时费力这一问题，探索从边缘检测的角度实现对地块边缘的自动提取。在构建耕地地块边缘遥感影像数据集工作中，尝试深度学习边缘检测模型holistically-nested edge detection（HED）和richer convolutional features（RCF）的基础上，进一步改变模型特征融合方式，并采用空洞卷积结构，提出构建应用于遥感影像的边缘检测模型full dilated-RCF（FD-RCF），提取耕地地块边缘。实验表明，相关方法的精度评定F1值均能达到0.8以上。构建的FD-RCF模型表现最佳，其检测结果在ODS和OIS精度评定中F1值分别达到0.848 1和0.850 2，平均精度0.795 7。比较而言，FD-RCF方法检测结果画面更加清晰，能够显著提高田间地形数据的更新效率。
High spatial resolution (high resolution) remote sensing images are reliable data sources for creating and updating field graphics databases. However, manual vectorization is a tedious process which costs much time and effort. In order to solve this problem, a method called full dilated-RCF (FD-RCF), which is based on dilated convolution in deep learning, is proposed. Compared with holistically-nested edge detection (HED) and richer convolutional features (RCF), FD-RCF mainly differentiates in the way of combining different results from different layers. FD-RCF consists of 6 stages and takes more care of the way of fusing these outputs from convolution layers into side-outputs. With dilated convolution, FD-RCF decreases the loss of information in deep layer. These methods can totally be used in detecting the boundary of agricultural fields. All of them get F1-values of over 0.8 in ODS and OIS. FD-RCF gets the highest F1-values of 0.848 1 and 0.850 2 in ODS and OIS, respectively, and the average precision of 0.795 7. The results gotten from FD-RCF are clearer than other methods and FD-RCF costs less time than manual vectorization.
[1] 李德仁.论21世纪遥感与GIS的发展[J].武汉大学学报(信息科学版), 2003, 28(2):127-131.
[2] 杜凤兰,田庆久,夏学齐,等.面向对象的地物分类法分析与评价[J].遥感技术与应用, 2004, 19(1):20-23.
[3] 宫鹏,黎夏,徐冰.高分辨率影像解译理论与应用方法中的一些研究问题[J].遥感学报, 2006, 10(1):1-5.
[4] Yang M D, Huang K S, Kuo Y H, et al. Spatial and spectral hybrid image classification for rice lodging assessment through UAV imagery[J]. Remote Sensing, 2017, 9(6):583.
[5] Liu W, Dong J, Xiang K, et al. A sub-pixel method for estimating planting fraction of paddy rice in Northeast China[J]. Remote Sensing of Environment, 2018, 205:305-314.
[6] Ryherd S, Woodcock C. Combining spectral and texture data in the segmentation of remotely sensed images[J]. Photogrammetric Engineering and Remote Sensing, 1996, 62(2):181-194.
[7] Mitra P, Shankar B U, Pal S K. Segmentation of multispectral remote sensing images using active support vector machines[J]. Pattern Recognition Letters, 2004, 25(9):1067-1074.
[8] Dey V, Zhang Y, Zhong M. A review on image segmentation techniques with remote sensing perspective[C]//Wagner W, Székely B. ISPRS TC VII Symposium-100 Years ISPRS. Vienna:IAPRS, 2010:31-42.
[9] Rydberg A, Borgefors G. Integrated method for boundary delineation of agricultural fields in multispectral satellite images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2001, 39(11):2514-2520.
[10] 胡潭高, 朱文泉, 阳小琼, 等. 高分辨率遥感图像耕地地块提取方法研究[J]. 光谱学与光谱分析, 2009, 29(10):2703-2707.
[11] Martin D R, Fowlkes C C, Malik J. Learning to detect natural image boundaries using local brightness, color, and texture cues[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2004(5):530-549.
[12] Arbelaez P, Maire M, Fowlkes C, et al. Contour detection and hierarchical image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(5):898-916.
[13] Lim J J, Dollar P, Zitnick III C L. Learned mid-level representation for contour and object detection:U.S. Patent Application 13/794,857[P]. (2014-09-18)[2018-11-10].
[14] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. arXiv preprint arXiv:1409.1556, 2014.
[15] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks[C]//Advances in Neural Information Processing Systems. 2012:1097-1105.
[16] Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015:1-9.
[17] Ren S, He K, Girshick R, et al. Faster r-cnn:Towards real-time object detection with region proposal networks[C]//Advances in Neural Information Processing Systems. 2015:91-99.
[18] Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015:3431-3440.
[19] Schulz H, Behnke S. Learning object-class segmentation with convolutional neural networks[C]//ESANN. 2012:151-156.
[20] Chen L C, Papandreou G, Kokkinos I, et al. Deeplab:Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(4):834-848.
[21] Xie S, Tu Z. Holistically-nested edge detection[C]//Proceedings of the IEEE International Conference on Computer Vision, 2015:1395-1403.
[22] Liu Y, Cheng M M, Hu X, et al. Richer convolutional features for edge detection[C]//Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on. IEEE, 2017:5872-5881.
[23] Dai J, Li Y, He K, et al. R-fcn:Object detection via region-based fully convolutional networks[C]//Advances in Neural Information Processing Systems, 2016:379-387.
[24] Hu F, Xia G S, Hu J, et al. Transferring deep convolutional neural networks for the scene classification of high-resolution remote sensing imagery[J]. Remote Sensing, 2015, 7(11):14680-14707.
[25] Cheng G, Han J, Lu X. Remote sensing image scene classification:benchmark and state of the art[J]. Proceedings of the IEEE, 2017, 105(10):1865-1883.
