Adaptive noise reduction method based on a modified Lee filter for SAR image classification tasks

The study aimed to create a set of software tools for automated processing and classification of synthetic aperture radar images using adaptive image analysis algorithms. The study used archival data from Sentinel-1, TerraSAR-X and RADARSAT-2 radar satellites and applies both classical image processing methods and adaptive algorithms. The quality of filtering, segmentation, classification, and object detection was assessed in terms of accuracy, structural similarity, signal-to-noise ratio, and consistency of results. The architecture of the software package was developed, including modules for pre-processing Synthetic Aperture Radar data, adaptive spectral filtering, image segmentation, and object classification. The study implemented adaptive algorithms such as the Lee filter, the K-means variant, the support vector method and the Ordered Statistics Constant False Alarm Rate. The developed tools were tested on satellite images from Sentinel-1 and RADARSAT-2 platforms for different types of the Earth’s surface. The adaptive filtering algorithm improved image quality by 35%, and performance on key metrics increased by 15-45% compared to traditional methods. High classification accuracy, including Kappa coefficient, F1, and area under the Receiver Operating Characteristic curve (Area Under the Curve), while maintaining computational efficiency, was provided. Automatic detection of water bodies, urban areas and agricultural land was implemented with an image processing time of less than 3 minutes. Adaptive algorithms ensured stable operation in conditions of different input data quality, making them suitable for a wide range of practical applications in the field of remote sensing and geographic information systems