Software package for processing and analysis of satellite SAR images
Radar images (SAR – Synthetic-Aperture Radar, synthetic aperture) can be obtained from satellites regardless of the weather and day time. Currently, modern satellites obtain images with a resolution of about 1 meter per pixel by means of aperture synthesis. Such images are well represented by human-made objects, especially the metal objects. However, images of this type have very strong spatial and luminance distortions. They are characterized by a high level of speckle structure arising as a result of coherent processing of the reflected signal, which has a random phase due to atmospheric fluctuations and the complex shape of the reflecting surface.
The tasks of processing SAR images:
geometric and radiometric correction;
analysis of the speckle structure to determine the statistical characteristics of reflective surfaces;
spatial frequency filtering, which provides selection and visualization of objects of a given type;
texture analysis, which provides a stable separation of various objects, taking into account the relief characteristics;
Scientific and technical groundwork.
Since 2010, the UIIP NAS of Belarus has been carrying out fundamental research on the processing of SAR images (task KI 1.27 "Development of algorithms and software complex for processing high-resolution radar satellite images" of SCRP "Space Research" for 2010–2012 (executive manager – A. Dudkin) and task 3.2.07 CI-2015 "To develop methods, algorithms and experimental programs for the operational processing of high-resolution radar satellite images based on geoinformation systems" (executive manager – V. Staravojtaŭ).
The following main results were obtained:
- algorithms for the preprocessing of radar satellite images, ensuring the selection and visualization of objects of a given type;
- algorithms for fuzzy clustering of radar satellite images based on the use of fractal and texture characteristics;
- algorithms for recognizing the types of areal objects based on the neural network classification of histogram characteristics;
- software complex for processing high-resolution radar satellite images, based on the developed processing algorithms (filtering, segmentation, classification), designed to solve problems of monitoring the state of areal objects based on the analysis of radar data of Earth surface remote sensing;
- filtering algorithms and methods for assessing the filtering quality of synthetic aperture radar images.
The method for adaptive compression of a wide dynamic range of radar space images with maximum preservation of image contrast has been developed as part of the contract No. F16SRBG-004 from July 20, 2016 (V. Staravojtaŭ).
Also, the systems for processing radar and optical signals, automatic recognition and selection of targets in real time and modeling broadband space-time radar signals using SKIF computational modules have been developed jointly with BSUIR as part of the STP of US "Development and implementing in serial production of a family of high-performance computing systems with parallel architecture (supercomputers) and the creation of applied software and hardware systems based on them" (Code "Skif", executive manager – L. Padzionak). The laboratory performed:
- selection of the type and parameters of the spatio-temporal structure of the probing signal (together with SOLKOMtech (S. Hiejster);
- development of the system for modeling broadband space-time radar signals;
- modeling the process of forming a radar signal reflected from a moving object of arbitrary shape (modeling the process of forming a radar portrait) when it is irradiated with a broadband signal on the Skif cluster.
The ensembles of neural networks for the classification of radar images have been developed as a part of pre-design studies for this project (participation in the international competition Statoil/C-CORE Iceberg> Classifier Challenge).
