Object-based Classification of Earthquake Damage from High-resolution Optical Imagery Using Machine Learning

Object-based approaches to the segmentation and supervised classification of remotely-sensed images yield more promising results compared to traditional pixelbased approaches. However, the development of an object-based approach presents challenges in terms of algorithm selection and parameter tuning. Subjective methods and trial and error are often used, but time consuming and yield less than optimal results. Objective methods are warranted, especially for rapid deployment in time sensitive applications such as earthquake induced damage assessment. Our research takes a systematic approach to evaluating object-based image segmentation and machine learning algorithms for the classification of earthquake damage in remotely-sensed imagery using Trimble’s eCognition software. We tested a variety of algorithms and parameters on post-event aerial imagery of the 2011 earthquake in Christchurch, New Zealand. Parameters and methods are adjusted and results compared against manually selected test cases representing different classifications used. In doing so, we can evaluate the effectiveness of the segmentation and classification of buildings, earthquake damage, vegetation, vehicles and paved areas, and compare different levels of multi-step image segmentations. Specific methods and parameters explored include classification hierarchies, object selection strategies, and multilevel segmentation strategies. This systematic approach to object-based image classification is used to develop a classifier that is then compared against current pixel-based classification methods for post-event imagery of earthquake damage. Our results show a measurable improvement against established pixel-based methods as well as object-based methods for classifying earthquake damage in high resolution, post-event imagery.