ABSTRACT
As the typical man-made objects, buildings play a special role in remote sensing images. This paper proposes a building scene recognition method based on deep multiple instance convolutional neural network. Scene classification is formulated as a multiple instance learning (MIL) problem in which local scene regions are regarded as instances and assigned with different labels. Through a trainable MIL pooling based on a spatial attention mechanism to select the most relevant instances adaptively and produce the scene-level predictions. Experimental results on UCM data set and small targets data set collected based on the GF-2 show that the method can improve the accuracy of building scene recognition using high resolution remote sensing image. Particularly, it solves the problem that it is difficult to detect the small buildings correctly and provide a new idea of scene recognition of buildings.
- Cao Q. Building extraction based on high resolution remote sensing image [D]. 2018. Jilin University.Google Scholar
- Han J, Zhang D, Cheng G, et al. 2015. Object detection in optical remote sensing images based on weakly supervised learning and high-level feature learning[J].IEEE Transactions on Geoscience and Remote Seinsing, 2015, 53(6):3325--3337.Google ScholarCross Ref
- Wu Z, Hu Z, Zhang Q, Cui W. 2013.On Combining Spectral, Textural and Shape Features for Remote Sensing Image Segmentation[J]. Acta Gesodaetica et Cartographica Sinica, 2013, 42(01):44--50.Google Scholar
- Cha Y, Ni Z, Yang S.2003.An Effective Approach to Automatically Extract Urban Land-use from TM Imagery[J]. Journal of remote sensing, 2003, 7(1):37--40.Google Scholar
- Lhomme S, He D C, Weber C, et al. A new approach to building identification from very high spatial-resolution images [J]. 2009, 30(5):1341--1354.Google ScholarDigital Library
- Yang Z, He X.2010.On Combining Spectral, Textural and Shape Features for Remote Sensing Image Segmentation [J]. Journal of Hohai University(Natural Sciences), 2010, 38( 2): 181--184.Google Scholar
- Wang J, Qin Q M, CHEN L, et al.Automatic Building Extraction from very High Resolution Satellite Imagery Using Line Segment Detector[C]// Geoscience and Remote Sensing Symposium ( IGARSS). [S.l.]: IEEE, 2013.Google Scholar
- Jung C R, Schramm R. Rectangle detection based on a windowed Hough transform[J].Computer Society, IEEE, 2004( 9): 113--120.Google ScholarCross Ref
- Borelli M, Ughi M.The fast diffusion equation with strong absorption: the instantaneous shrinking phenomenon [J]. Rend Istit Mat Univ Trieste, 1994, 26: 109--140.Google Scholar
- Zhang J, Marszalek M, Lazebnik S, et al. Local Features and Kernels for Classification of Texture and Object Categories: A Comprehensive Study[J].International Journal of Computer Vision, 2007, 73(2):213--238.Google ScholarDigital Library
- Vakalopoulou M, Karantzalos K, Komodakis N, etal.Building detection in very high resolution multispectral data with deep learning features[C]//Geoscience and Remote Sensing Symposium, IEEE, 2015: 1873--1876.Google Scholar
- Li Zhili, Xu Kai, Han Wenjun, Bi Qi, Qin Kun. Deep Multiple Instance Learning for Scene Classification Using Remote Sensing Images [C]. 2018--10th IAPR Workshop on PRRS, 19-20 August 2018, Beijing, China.Google Scholar
- Dietterich T G, Lathrop R H, Lozano-Perez T. Solving the multiple instance problem with axis-parallel Rectangles [J]. Artificial intelligence, 1997, 89(1-2):31--71.Google ScholarDigital Library
- Krizhevsky A, Sutskever I, Hinton G.ImageNet Classification with Deep Convolutional Neural Networks[C]// NIPS. Curran Associates Inc. 2012.Google Scholar
- Xia G S, Hu J, Hu F, et al. AID: A benchmark data set for performance evaluation of aerial scene classification. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(7): 3965--3981.Google ScholarCross Ref
Index Terms
- Building Scene Recognition Based on Deep Multiple Instance Learning Convolutional Neural Network Using High Resolution Remote Sensing Image
Recommendations
Recognition and extraction of high-resolution satellite remote sensing image buildings based on deep learning
AbstractExtracting and recognizing buildings from high-resolution remote sensing images faces many problems due to the complexity of the buildings on the surface. The purpose is to improve the recognition and extraction capabilities of remote sensing ...
Microscopic image super resolution using deep convolutional neural networks
AbstractRecently, deep convolutional neural networks (CNNs) have achieved excellent results in single image super resolution (SISR). Owing to the strength of deep CNNs, it gives promising results compared to state-of-the-art learning based models on ...
TQR-Net: Tighter Quadrangle-Based Convolutional Neural Network for Dense Building Instance Localization in Remote Sensing Imagery
Image and GraphicsAbstractBuilding localization in remote sensing imagery (RSI) is widely applied in many geoscience and remote sensing areas. However, many existing methods cannot generate accurate building contours. In this paper, we propose an effective convolutional ...
Comments