Extracting Building Footprints from Satellite Images using Convolutional Neural Networks

Chawda, Chandan; Aghav, Jagannath; Udar, Swapnil

doi:10.1109/icacci.2018.8554893

Cited by 6 publications

(4 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Footprint extraction models We make a reference to building footprint extraction models because these models aim to extract (either via semantic or instance segmentation) complex geometries that come from a much more complex distribution space -in terms of shape and texture variance -than features such as vehicles or trees [3,6,32]. Building footprints are also a class of features which are conditionally dependent on contextual information; a building is much more likely to be found in a suburban context next to trees or driveways as opposed to in the desert.…”

Section: Related Workmentioning

confidence: 99%

“…Existing approaches however rely heavily on segmentation labels [6,32] and primarily focus on improving the accuracy from a supervised learning approach given large amounts of training data, such as in the SpaceNet [11] and WHU [29] datasets. This is unfortunately not readily available for many worthwhile segmentation tasks such as well pad extraction for localizing new fracking operations or airplane detection for civil and remote sensing applications.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Points2Polygons: Context-Based Segmentation from Weak Labels Using Adversarial Networks

Yu,

Frank,

Wilson

2021

Preprint

View full text Add to dashboard Cite

In applied image segmentation tasks, the ability to provide numerous and precise labels for training is paramount to the accuracy of the model at inference time. However, this overhead is often neglected, and recently proposed segmentation architectures rely heavily on the availability and fidelity of ground truth labels to achieve state-of-the-art accuracies. Failure to acknowledge the difficulty in creating adequate ground truths can lead to an over-reliance on pre-trained models or a lack of adoption in real-world applications. We introduce Points2Polygons (P2P), a model which makes use of contextual metric learning techniques that directly addresses this problem. Points2Polygons performs well against existing fully-supervised segmentation baselines with limited training data, despite using lightweight segmentation models (U-Net with a ResNet18 backbone) and having access to only weak labels in the form of object centroids and no pre-training. We demonstrate this on several different small but non-trivial datasets. We show that metric learning using contextual data provides key insights for self-supervised tasks in general, and allow segmentation models to easily generalize across traditionally label-intensive domains in computer vision.Preprint. Under review.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Points2Polygons: Context-Based Segmentation from Weak Labels Using Adversarial Networks

Yu,

Frank,

Wilson

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Although these algorithms have strong building detection capabilities, their performance depends heavily on the quantity and quality of the training datasets (Roscher et al, 2020, Chen et al, 2023. Addressing the issue of limited building label samples, (Chawda et al, 2018) manually created labels and images from Bing Maps to train the model. Topographic maps have been utilized to create building samples for very-high-resolution (VHR) Earth observation (EO) remote sensing images.…”

Section: Introductionmentioning

confidence: 99%

A Large-Scale Multipurpose Benchmark Dataset and Real-Time Interpretation Platform Based on Chinese Rural Buildings

Deng,

Xu,

Huang

et al. 2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

As urbanization accelerates, the evolving dynamics of village growth and decline have garnered widespread attention. Rural housing, as the most significant asset in villages, serves as the primary indicator of socio-economic development in rural areas. However, the extensive scale, diversity, and widespread distribution of villages make conducting a nationwide census of rural buildings a notably costly and time-intensive endeavor. Although deep learning techniques have been successfully applied by numerous researchers to map building footprints, the majority of this work is concentrated in urban areas, leaving large-scale datasets for rural buildings notably lacking. In this article, an exhaustive database of rural architecture has been established, featuring diverse rural building annotations from the majority of provinces in mainland China. Moreover, a realtime online platform for remote sensing image interpretation, integrating instance segmentation and boundary regularization, has been developed to streamline the extraction of building footprints from high-resolution imagery. Experimental results from predicting 43,992 rural building instances nationwide demonstrated that 33,210 were accurately identified, achieving a precision of 0.776, a recall of 0.755, and an F1 score of 0.765. Building upon this work, the maps of rural building areas and quantity are produced to clearly demonstrate the distribution of rural houses in parts of China. These data products can serve as vital supplements to public data products such as nighttime light data, land cover maps, national statistical yearbooks, and road network data, particularly in the field of rural studies.

show abstract

“…Various authors have reported studies for transitioned buildings footprints identification using post classification comparison of temporal classified images [16,17]. However, post classification change detection approach is depended on the accuracy of individual date classification outputs.…”

Section: Introductionmentioning

confidence: 99%

Fuzzy machine learning approach for transitioned building footprints extraction using dual-sensor temporal data

2021

View full text Add to dashboard Cite

This study presents a fuzzy approach, for detection of transitioned building footprints in urban area using medium resolution datasets. Multi-temporal remote sensing data sets from Landsat-8 Operational Land Imager and Sentinel-2A were used for generation of temporal indices database. The database was generated using class-based sensor independent-normalized difference vegetation index approach, with an aim to reduce spectral dimensionality of each image and maintain temporal dimensionality. The temporal indices database was subsequently used as input in Modified Possibilistic c-means classifier for transitioned building footprints extraction. The identified transitioned building locations were validated using ground samples as well as from Google images at four different test sites. For accuracy assessment, F-measure was calculated and its value was 0.75 or higher for all training and testing sites. Thus, using proposed fuzzy approach, transitioned building footprints were accurately identified compared to traditional techniques.

show abstract

Extracting Building Footprints from Satellite Images using Convolutional Neural Networks

Cited by 6 publications

References 20 publications

Points2Polygons: Context-Based Segmentation from Weak Labels Using Adversarial Networks

Points2Polygons: Context-Based Segmentation from Weak Labels Using Adversarial Networks

A Large-Scale Multipurpose Benchmark Dataset and Real-Time Interpretation Platform Based on Chinese Rural Buildings

Fuzzy machine learning approach for transitioned building footprints extraction using dual-sensor temporal data

Contact Info

Product

Resources

About