Weakly Supervised Road Segmentation in High-Resolution Remote Sensing Images Using Point Annotations

Road segmentation methods based on deep neural networks have achieved great success in recent years, but creating accurate pixel-wise training labels is still a boring and expensive task, especially for large-scale high-resolution remote sensing images (HRSIs). Inspired by the stacked hourglass model for human joints detection, we propose a weakly supervised road segmentation method using point annotations in this article. First, we design a patch-based deep convolutional neural network (DCNN) model for road seeds and background points detection and train the model using point annotations. Then, in the process of road segmentation, the DCNN model detects a series of road and background points that are used to train a Support Vector Machine Classifier (SVC) for classifying each pixel into road or nonroad. According to the local geometry of road and the inaccurate classification of SVC, a multiscale and multidirection Gabor filter (MMGF) is put forward to estimate the road potential. Finally, the active contour model based on local binary fitting energy (LBF-Snake) is introduced to extract the road regions from the inhomogeneous road potential. Qualitative and quantitative comparisons show that our method achieves results close to the fully supervised semantic methods without considering the annotation cost and outperforms them given a fixed budget.

I. Introduction

Road segmentation in the remote sensing images is the hot research issue due to its vital applications in cartography, traffic planning, vehicle navigation, intelligent transportation system, and so on. Despite a huge number of works dealing with this problem, the results are still limited and not generic enough to solve the lack of information due to the existence of artifacts such as overcast of building shadows, roadside trees, or cars [1]. With the development of computer vision, deep learning technologies have been widely used in the field of road segmentation. However, training these deep learning models demands massive training samples, while annotating such a large number of samples is a tough, boring, and expensive task. In order to reduce the label cost of training data, weakly supervised semantic segmentation has become a research hotspot in recent years.