CI-dataset and DetDSCI methodology for detecting too small and too large critical infrastructures in satellite images: Airports and electrical substations as case study

The detection of critical infrastructures in large territories represented by aerial and satellite images is high importance several fields such as security, anomaly detection, land use planning change detection. However, the complex they have highly variable shapes sizes, i.e., some infrastructures, electrical substations, are too small while others, airports, large. Besides, airports can a surface area either or with completely different shapes, which makes its correct challenging. As far we know, these limitations not been tackled yet previous works. This paper presents (1) smart Critical Infrastructure dataset, named CI-dataset, organized into two scales, scales (2) two-level resolution-independent infrastructure (DetDSCI) methodology that first determines spatial resolution input image using classification model, then analyses appropriate detector for resolution. present study targets representative classes, substations. Our experiments show DetDSCI achieves up to 37.53% F1 improvement respect Faster R-CNN, one most influential models.