Road Segmentation Based on Hybrid Convolutional Network for High-Resolution Visible Remote Sensing Image
This paper aim to detect the roads in aerial images is taken by satellite images.
A hybrid convolution network (HCN) to segment roads from high-resolution visible remote sensing images, with the hope to achieve high segmentation for multi scale roads with complex background. The FCN has a few false segmentations on complex background and segments most of the wide roads, but it misses lots of the narrow roads and fails to segment the accurate boundaries of the roads. The poor performance of FCN on narrow roads is due to its lack of road details during the down sampling to up sampling operation.
In this proposed system the method of image contrast enhancement on the basis of the contrast distribution at the boundaries of objects and background on the image is proposed. Besides, deep learning methods especially neural network have also been widely used for road recognition and segmentation.