Object Level Change Detection Using Deep Learning Techniques

Change detection from remotely sensed imagery is critical for many applications, including land use mapping. In recent years, an increasing number of researchers have applied capable deep learning methods to change detection research. The vast majority of deep learning-based change detection methods currently perform pixel-by-pixel classification at the original image scale, but they are hardly immune to the problem. False changes caused by strong parallax effects and projected shadows, without taking the totality of changed objects/regions into account. In this paper, we propose an object-level change detection framework for detecting changed geographic entities (such as newly constructed buildings or changed artificial structures) by focusing on the overall characteristics and context association of changed object instances. The detected changed objects are represented as bounding boxes, which are simple, regular, and useful for extracting object features. .

Keywords: Object level image data, Segmentation, Change detection