Road Extraction from Multi-Source High-Resolution Remote Sensing Image Using Convolutional Neural Network

摘要

The traditional network information update depends on the field measurement or artificial interpretation of surveying and mapping in remote sensing image. The shortcomings are obvious: high cost, long cycle, and consume a large amount of manpower. In order to solve this problem, we use the convolution neural network in the deep learning to complete road information extraction from high resolution image. The road training database combines two kinds of high resolution remote sensing data which are the GaoFen-2 and World View. Two different training models are used to compare the results. Furthermore, the results of the two models are combined to obtain more accurate and improved road extraction results.