Integrated approaches for road extraction and de-noising in satellite imagery using probability neural networks

Abstract

Remote sensing plays a crucial role in detecting and monitoring natural resources, extending its applications to various fields, such as geography, topographical surveying, and geoscience disciplines, including land management, forest monitoring, crop identification, soil mapping, and ocean resource finding. Road extraction holds significant importance among these applications, contributing to the development of Geographic Information Systems (GIS). The automatic updating of GIS information has become essential in daily life. Road extraction stands out as a prominent application within remote sensing image systems, addressing challenges related to intensity and width. Intensity challenges involve variations in spectral or color values of roads, while width challenges pertain to the issues associated with the size and structure of roads during the extraction process. Addressing the challenges associated with road extraction from remote sensing imagery is crucial for achieving accurate and efficient results. This paper under consideration compares the conventional and contemporary methods of road extraction, emphasizing completeness and correctness metrics. Conventional methods involve techniques like CLAHistogramEqualization for enhancement and fuzzy c-mean clustering for extraction, resulting in incremental improvements. To enhance results further, images are de-noised using Gray World Optimization and iterative domain-guided image filtering. To improve efficiency in road extraction, the authors proposed a contemporary approach through probability neural networks with de-noised images. The comparison is made based on the completeness and correctness of both conventional and contemporary methods.