Image-Based Pre- and Post-conditional Probability Learning for Efficient Situational Assessment and Awareness

With the increasing amount and severity of environmental disasters, the importance of situational assessment and awareness for human assistance and disaster response is critical. During natural disasters in populated regions, proper assistance and response efforts can only be planned and deployed effectively when the damage levels can be resolved promptly. Damage-level assessment is aided by aerial imagery. While human labeling of images provides a measure of credibility, in the presence of a large volume of image data, it takes a long time and great effort to achieve situational assessment and awareness which can significantly hamper the operational response time. Recently, extreme computational power has enabled Deep LearningDeep learning to analyze large images. This paper presents an efficient and scalable humanitarian assistance and disaster response application for situational assessment and awareness using a method called Image-based Pre- and Post-conditional Probability learning, which matches the pre- and post-disaster images by effectively encoding one image that determines the damage levels through a deep learningDeep learning method. Two example scenarios of humanitarian assistance and disaster response applications are examined: (1) pixel-wise semantic segmentation, and (2) contrastive learning patch-based damage classificationClassification, both showing promising results in the examined scenarios which motivates the application of the deep learningDeep learning enhanced methods based on our new method to achieve computational efficiency while maintaining classification accuracy.