Learning to Detect Roads in High-Resolution Aerial Images

By Volodymyr Mnih, Geoffrey E. Hinton. European Conference on Computer Vision, 2010

663 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Environmental Engineering, Ocean Engineering

View Paper: https://www.semanticscholar.org/paper/8b16106e35ff9ff97ffb73eb9c84b73ce1264c67

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Environmental Engineering

In Ocean Engineering

Citations in the corpus

Context-Dependent Pre-Trained Deep Neural Networks for Large-Vocabulary Speech Recognition (2012, background): “…deep architectures than deep autoencoders can in some cases be trained without pre-training, for many problems and model architectures, researchers have reported pre-training to be helpful (even in some cases for large single hidden layer neural networks trained on massive datasets, as in [28]).”
Self-Driving Cars: A Survey (2019, methodology): “For reviews on methods for traffic sign detection and recognition, readers are referred to Mogelmose et al. (2012) and Gudigar et al.”
Transfer Learning from Deep Features for Remote Sensing and Poverty Mapping (2015, background): “Amazingly, these features are learned without direct supervision, in contrast to previous efforts to extract features from aerial imagery, which have relied heavily on large amounts of expert-labeled data, e.g., labeled examples of roads (Mnih and Hinton 2010; 2012).”
Machine Learning for Aerial Image Labeling (2013, methodology): “…of large-scale machine learning to such high-resolution imagery have produced object detectors with impressive levels of accuracy [Kluckner and Bischof, 2009, Kluckner et al., 2009, Mnih and Hinton, 2010, 2012], suggesting that automated aerial image interpretation systems may be within reach.”
Fully Convolutional Networks for Dense Semantic Labelling of High-Resolution Aerial Imagery (2016, background): “[20, 19] trained CNNs to detect roads and buildings from automatically-generated noisy ground truth labels.”
Semantic Segmentation of Earth Observation Data Using Multimodal and Multi-scale Deep Networks (2016, methodology): “Since the first works on road detection [17], CNN have been successfully used for classification and dense labeling of EO data.”
DeepRoadMapper: Extracting Road Topology from Aerial Images (2017, methodology): “Convolutional neural networks have been used to segment roads from aerial images [15].”
Classification and Segmentation of Satellite Orthoimagery Using Convolutional Neural Networks (2016, background): “Contextual information has also been utilized to learn to differentiate roads from other classes using neural networks [41,42].”
Deep Learning Classification of Land Cover and Crop Types Using Remote Sensing Data (2017, background): “land cover types such as road extraction, buildings extraction [17], [27], [28].”
Satellite Imagery Classification Based on Deep Convolution Network (2016, methodology): “Mnih and Hinton [2] proposed a method that uses Deep Neural Networks to detect roads in Aerial imagery.”
Learning to Label Aerial Images from Noisy Data (2012, methodology): “The best published results on this data (Mnih & Hinton , 2010) make use of a postprocessing procedure that improves the predictions of a base model by training a new predictor that takes a patch of predictions of the base model as input instead of the aerial image.”
DeepSat: a learning framework for satellite imagery (2015, methodology): “Also, in [26], the authors used a window of size 64×64 to derive contextual information.”
D-LinkNet: LinkNet with Pretrained Encoder and Dilated Convolution for High Resolution Satellite Imagery Road Extraction (2018, background): “Mnih and Hinton [17] employed restricted Boltzmann machines to segment road from high resolution aerial images.”
Learning Aerial Image Segmentation From Online Maps (2017, background): “We note that (nonconvolutional) deep networks in conjunction with OSM labels have also been applied for patch-based road extraction in overhead images of ≈ 1 m GSD at large scale [46], [47].”
Multitask learning for large-scale semantic change detection (2018, methodology): “For examples, in remote sensing, CNNs were used for road detection (Mnih and Hinton, 2010), and in computer vision, CNNs were used on the related task of comparing image pairs (Chopra et al.”
Classification With an Edge: Improving Semantic Image Segmentation with Boundary Detection (2016, background): “Mou and Zhu (2016) prefer to avoid upsampling altogether, 218 and instead combine the coarser semantic segmentation with a super-pixel 219 segmentation of the input image to restore accurate segmentation boundaries 220 (but not small objects below the scale of the fcn output). Of course, such 221 a ”