PROJECT TITLE :

Toward a Generalizable Image Representation for Large-Scale Change Detection: Application to Generic Damage Analysis

ABSTRACT:

Each year, multiple catastrophic events impact vulnerable populations around the earth. Assessing the harm caused by these events during a timely and accurate manner is crucial for efficient execution of relief efforts to assist the victims of those calamities. Given the low accessibility of the broken areas, high-resolution optical satellite imagery has emerged as a valuable source of knowledge to quickly asses the extent of injury by manually analyzing the pre- and postevent imagery of the region. To create this analysis more economical, multiple learning techniques using a selection of image representations are proposed. But, most of those representations are susceptible to variabilities in capture angle, sun location, and seasonal differences. To evaluate these representations within the context of damage detection, we tend to gift a benchmark of 86 pre- and postevent image pairs with respective reference data derived from United Nation Operational Satellite Applications Programme (UNOSAT) assessment maps, spanning a complete space of 4665 km2 from 11 completely different locations around the world. The technical contribution of our work may be a novel image illustration primarily based on form distributions of image patches encoded with locality-constrained linear coding. We tend to empirically demonstrate that our proposed representation provides an improvement of a minimum of five%, in equal error rate, over alternate approaches. Finally, we have a tendency to present an intensive robustness analysis of the considered representational schemes, with respect to capture-angle variabilities and multiple sensor combinations.