Naumenko O. SAR-images processing methods using artifitial neural networks

The research object - the remote sensing image processing that are corrupted by the intense noise; the research goal - is to increase the efficiency of the remote sensing image preprocessing methods (compression and adaptive filtering) by improving the local activity detection methods as well as methods for detection efficiency evaluation; the research methods - methods of probability theory and math statistics, numerical modeling, theory of nonlinear and adaptive filtering, spectral correlation analysis, machine learning methonds; practical results - increased efficiency of detecting inhomogeneities on SAR images by using neural network-based detector, improved the efficiency of the locally adaptive filtering using neural network-based local activity indicator; shown the versatility of the method for different types of noise, improved the efficiency of SAR-image filtering prediction method; novelty - the use of a neural network for combining several local heterogeneity detectors has been proposed for the first time, the two local heterogeneity detectors were improved for their application in the presence of various types of noise, the local heterogeneity detector on the basis of a discrete cosine transform has been proposed for the first time for detecting inhomogeneities in the spectral region, the classifier specificity and sensitivity efficiency estimation method has been improved and adapted for the automation of classifier training, method of filtering efficiency prediction have been improved; degree of implementation - results have been implemented in CASRE IGS NAS of Ukraine; area of use - remote sensing.