Intelligent Interpreting System of High Resolution Remote Sensing Image Based on Convolutional Network and Super Vector coding

摘要

The high resolution earth observation system project of China will coordinate the construction of high resolution geodetic systems based on satellites, stratospheric airships and aircraft, forming an all-weather, all-day, global coverage of ground-based observing capabilities. Intelligent interpretation of remote sensing images refers to the image detection, recognition and classification using computers, determining the object attributes or characteristics of the images. And then analyses and describes the various structures and relationships in the image, and explains the attributes, categories and relationships of the images, explains the variation of the various objects in time and space and the correspondence between them. In this paper, we propose a small object detection method based on deep Residual Network and Super-Vector coding. In our method, a variation of ResNet with fewer layers is defined to increase the resolution of the feature map and multi-level convolutional features are merged into an informative feature description for region proposal. Meanwhile, we extract histogram of oriented gradient (HOG) with SV coding from each region of interest, which assist convolutional features to complete object classification. We comprehensively evaluate the proposed method on our small-size object dataset. The experimental results show that our method outperforms top-performing methods on small object detection with higher accuracy and less time consuming.