Deep Built-Structures Counting In Satellite Imagery Using Attention Based Re-weighting

 

 

Abstract

The paper addresses the challenging problem of counting built-structures in the satellite imagery. Building density is a more accurate estimate of the population density, urban area expansion and its impact on the environment, than the built-up area segmentation. However, building shape variances, overlapping boundaries, and variant densities make this a complex task. A deep learning based regression technique for counting built-structures is proposed. To train and evaluate the method, we put forward a new large-scale and challenging built-structure-count dataset. Our dataset is constructed by collecting satellite imagery from diverse geographical areas (planes, urban centers, deserts, etc.,) across the globe (Asia, Europe, North America, and Africa) and captures the wide density of built structures. Detailed experimental results and analysis validate the proposed technique. FusionNet has Mean Absolute Error of 3.65 and R-squared measure of 88% over the testing data. Finally, we perform the test on the 274.3 × 103 meter square of the unseen region, with the error of 19 buildings off the 656 buildings in that area.

 

Model Overview:

A Fusion network is designed that merges information captured from three branches, two of which are attention based where as the third is fully connected trained on DenseNet features. The Cross Chanel Parametric Pooling (CCPP) and Global Weighted average Pooling (GWAP) both are attention based and filters DenseNet features using the SS-NET (Satellite Segmentation Network). The two different ways of adding attention is fully explained in the Section 4 of our paper. Initially the two bracnhes are trained separately and then merged together in the Fusion Net. The system diagram of CCPP, GWAP and Fusion Net is shown below.

 

Dataset:

We have collected a large dataset of diverse regions from around the globe. Label for each image is total number of houses in it. The dataset can be download from here.

 

SSNet:

Satellite Segmentation Network calculates per pixel probability for built-up presence on an image.

Github Repository:

To download code and pre-trained models for FusionNet, please visit our github repository.

Other Links:

You can find our publication here.

bibtex:

@article{shakeel2019deep,
  title={Deep built-structure counting in satellite imagery using attention based re-weighting},
  author={Shakeel, Anza and Sultani, Waqas and Ali, Mohsen},
  journal={ISPRS Journal of Photogrammetry and Remote Sensing},
  volume={151},
  pages={313--321},
  year={2019},
  publisher={Elsevier}
}