DeepWindow: Sliding Window Based on Deep Learning for Road Extraction From Remote Sensing Images

Abstract

The road centerline extraction is the key step of the road network extraction and modeling. The hand-craft feature engineering in the traditional road extraction methods is unstable, which makes the extracted road centerline deviated from the road center in complex cases and even results in overall extracting errors. Recently, the road centerline extraction methods based on semantic segmentation employing deep neural network greatly outperformed the traditional methods. Nevertheless, the pixel-wise labels for training deep learning models are expensive and the postprocess of road segmentation is error-prone. Inspired by the work of human pose estimation, we propose DeepWindow, a novel method to automatically extract the road network from remote sensing images. DeepWindow uses a sliding window guided by a CNN-based decision function to track the road network directly from the images without the prior of road segmentation. First of all, we design and train a CNN model to estimate the road center points inside a patch. Then, the road seeds are automatically searched patch by patch employing the CNN model. Finally, starting from seeds, our method first estimates the road direction using a Fourier spectrum analysis algorithm and then iteratively tracks the road center-line along the road direction guided by the CNN model. In our method, the CNN model is trained by point annotations, which greatly reduces the training costs comparing to those in semantic model training. Our method achieves comparable performance with the state-of-the-art road extraction methods, and extensive experiments indicate that our method is robust to the point deviation.

Road Extraction Methods in High-Resolution Remote Sensing Images: A Comprehensive Review

“…Bastani et al [140] proposed a new method called RoadTracer to extract accurate road network directly from aerial images using an iterative search process guided by a patch-based CNN decision function. Inspired by [142] and [140], Lian et al [143] presented a road network tracking algorithm. The key component is a road center points estimation DCNN model.…”

A Review of Remote Sensing Applications on Very High-Resolution Imagery Using Deep Learning-Based Semantic Segmentation Techniques

“…Regarding Semantic Segmentation [84,147,148,149], there are some use cases, such as building footprint extraction [11,12,150,13,14,15,16,17,18], road extraction [151,152,153] and land use and land cover (LULC) analysis [154,155]. • SpaceNet [158,159]: dataset with satellite imagery of the following cities: Rio de Janeiro, Las Vegas, Paris, Khartoum, and Shanghai.…”

Panoptic Segmentation Meets Remote Sensing

“…Second, the instance segmentation databases are usually monothematic, as many building footprints datasets such as the SpaceNet competition [37]. A good starting point for a large remote sensing dataset would include widely used and researched targets, and the urban setting and its components is a very hot topic with many applications: road extraction [38][39][40][41][42][43][44][45], building extraction [46][47][48][49][50][51][52], lake water bodies [53][54][55], vehicle detection [56][57][58], slum detection [59], plastic detection [60], among others. Most studies address a single target at a time (e.g., road extraction, buildings), and panoptic segmentation would enable vast semantic information of images.…”