Fully Convolutional Networks for Dense Semantic Labelling of High-Resolution Aerial Imagery

By Jamie Sherrah. arXiv.org, 2016

397 citations recorded by the source corpus.

The trend towards higher resolution remote sensing imagery facilitates a transition from land-use classification to object-level scene understanding. Rather than relying purely on spectral content, appearance-based image features come into play. In this work, deep convolutional neural networks (CNNs) are applied to semantic labelling of high-resolution remote sensing data. Recent advances in fully convolutional networks (FCNs) are adapted to overhead data and shown to be as effective as in other domains. A full-resolution labelling is inferred using a deep FCN with no downsampling, obviating the need for deconvolution or interpolation. To make better use of image features, a pre-trained CNN is fine-tuned on remote sensing data in a hybrid network context, resulting in superior results compared to a network trained from scratch. The proposed approach is applied to the problem of labelling high-resolution aerial imagery, where fine boundary detail is important. The dense labelling yields state-of-the-art accuracy for the ISPRS Vaihingen and Potsdam benchmark data sets.

Subfields: Environmental Engineering, Media Technology, Ocean Engineering

View Paper: https://www.semanticscholar.org/paper/83ef7de2669bb2827208fd3a64ac910e276fbdb4

How researchers see this paper

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

In Environmental Engineering

In Media Technology

In Ocean Engineering

Citations in the corpus

Land-cover classification with high-resolution remote sensing images using transferable deep models (2018, methodology): “To address this problem, an effective solution is to replace down-sampling process with the structure that preserves spatial information (Sherrah, 2016; Persello and Stein, 2017), such as the max pooling indices applied in SegNet (Badrinarayanan et al., 2017), and the dilated convolutions employed…”
PatternNet: A Benchmark Dataset for Performance Evaluation of Remote Sensing Image Retrieval (2017, background): “Inspired by the great success of deep learning, the remote sensing community has realized the potential of applying deep learning techniques for remote sensing tasks such as scene classification [34–38], object detection [39–41], semantic segmentation [42,43], image super-resolution [44] and image r”
Deep learning in remote sensing: a review (2017, methodology): “They train several CNNs, using them as inputs for the activation maps of each layer of the main model as well as one fusing the CIR and DSM main streams, as in [137].”
Dense Semantic Labeling of Subdecimeter Resolution Images With Convolutional Neural Networks (2016, methodology): “a few additional accuracy points, as pointed out in [36].”
Simultaneous extraction of roads and buildings in remote sensing imagery with convolutional neural networks (2017, background): “…Paisitkriangkrai et al., 2015; Kampffmeyer Michael and Jenssen, 2015; Jiang et al., 2015; Vakalopoulou et al., 2015; Castelluccio et al., 2015; Sherrah, 2016; Nogueira et al., 2016), but also in some networks that have been proposed in computer vision, e.g., fully connected network (Long et…”
Classification With an Edge: Improving Semantic Image Segmentation with Boundary Detection (2016, methodology): “57 There has been some research that tries to mitigate the blurring of bound58 aries due to down-sampling and subsequent up-sampling, either by using the 59 a-trous convolution (dilated convolution) (Yu and Koltun, 2016; Chen et al., 60 2016a; Sherrah, 2016) or by adding skip connections from early ”
Continual Learning With Structured Inheritance for Semantic Segmentation in Aerial Imagery (2021, methodology): “Following the prior works [56], we select seven images from training images as the validation set.”
Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art (2017, background): “Sherrah (2016) propose to use fully-convolutional net-”
Land cover mapping at very high resolution with rotation equivariant CNNs: towards small yet accurate models (2018, methodology): “In our setup, 11 of the 16 fully annotated image tiles are used for training, and the remaining ones (tile ID 11, 15, 28, 30, 34) for testing, as in Sherrah (2016); Volpi and Tuia (2017); Maggiori et al. (2017).”
The color out of space: learning self-supervised representations for Earth Observation imagery (2020, background): “Indeed, a synergy between Earth Observation and Deep Learning techniques led to promising results, as highlighted by recent advances in land use and land cover classification [6], image fusion [7], and semantic segmentation [8].”
Building Extraction in Very High Resolution Remote Sensing Imagery Using Deep Learning and Guided Filters (2018, methodology): “A deep FCN with no downsampling was introduced to boost the effective training sample size and improve the classification accuracy [37].”
Fully Convolutional Networks for Multisource Building Extraction From an Open Aerial and Satellite Imagery Data Set (2019, neutral): “Sherrah [35] found a pretrained CNN fine-tuned on remote sensing data can lead to better results compared to a network trained from scratch.”
Gated Convolutional Neural Network for Semantic Segmentation in High-Resolution Images (2017, background): “In addition, high-resolution images contain many objects with high intra-class variance and low inter-class variance [7,8].”