Remote Sensing Image Segmentation by Combining Feature Enhanced with Fully Convolutional Network

摘要

The main idea of this paper is the 25-categories classification task of remote sensing satellite image which is provided by Beijing AXIS Technology Company Limited, and proposes new methods based on fully convolutional network (FCN) and image processing. This method utilized image processing to realize color mapping and feature enhanced of remote sensing satellite image. Consider the influence of equipment and scene shooting environment, there are differences in color performance between remote sensing images, we use color mapping to improve color consistency. Aiming at the disadvantage of FCN has lower sensitivity to details, we add edge information into image as an important signal and expand the image into a five-dimensional one. Then the classification results will be attained through 25-categories classification according to FCN model. The experiment result showed the method is able to enhance the accuracy of FCN model classification to some extent.