Presentation concerns SATREC software that was developed by SATIM within the ESA project “Development of an innovative software called SATREC for identification of objects on satellite SAR imagery by modelling and recognition of radar signatures”. SATREC allows stakeholders, even not necessarily familiar with SAR technology, to conduct highly advanced analysis and interpretation of radar images in the context of objects recognition. This software is made flexible enough to be a stand-alone application and, in parallel, a batch processing software that can be incorporated into one of the ESA’s Thematic Exploitation Platforms and used by a wide range of users. The developed version of SATREC software consist of two key aspects: modelling synthetic SAR signatures for selected 3D objects (using SSIG module) and recognition of objects based on generated synthetic signatures on real SAR images (using OREC module). The SSIG module is based on a ray tracing technique, which simplifies wave propagation in medium by considering geometrical lines from the wave source to the receiver (forward method) or from the receiver to the source (backward method). Responses of the ray traversal through the virtual scene is interpreted to reproduce that scene as an image. The adaptation of ray tracing technique in SSIG requires a 3D scene and object of which SAR signature needs to be simulated. Therefore, SSIG uses a triangle-based 3D object, the most popular method of object representation for ray tracing technique. By using special programming techniques and GPU cards for computations, the generation of the database of synthetic signatures of selected 3D objects is performed in near real time. This way, the database of signatures is regenerated for each SAR image used for analysis, based on its incidence angle, pixel spacing and other parameters. This helps to significantly decrease the degrees of freedom in the recognition procedure. In the SAR signature simulation Phong reflection model was used. The OREC module implement the idea of objects recognition on satellite SAR imagery using the simulated SAR signatures of objects. The process involves matching those objects (understood as a selected region of a SAR image with an unknown object) with existing patterns from the established database. Object recognition on SAR images within OREC software is conducted based on comparison of the object with database patterns using template matching method. For this method to operate correctly, it is crucial to prepare a database of signatures, which contains all possible representations of analyzed objects. Then OREC recognize an object based on its comparison with patterns, and provide a feedback about patterns, which are visually the most similar and are available in a database. It also provides information about the similarity of analyzed object to the patterns from the database.
|ESA Technical Officer||Nicolas Floury|