Publication Date:
2020-02-12
Description:
This paper presents two concepts for the detection of different objects from high spatial resolution satellite imagery like MOMS-02. The first method is a supervised image segmentation technique which allows a target oriented search for smaller objects. Based on the variation of threshold parameters, different regions are formed by pixel aggregation and selected by a shape classification using artificial neural networks or template matching. In the second approach, large objects such as field structures are extracted. Previously developed recognition methods for aerial imagery are limited to the detection of objects with simple regular shapes like houses and streets. Other objects which include fine details and complex outlines cannot be extracted but are also of great interest for many applications. Therefore, a new hierarchical technique is introduced which deletes small details and preserves the essential object shapes. The obtained results confirm that shape based algorithms are capable of detecting different objects.
Keywords:
550 - Earth sciences
Type:
info:eu-repo/semantics/conferenceObject
