Satellite Image Analysis for Land Use Land Cover Applications

Abstract

Recently, very high-resolution (VHR) remote-sensing images that have become more newlinepopular due to improved image quality and resolutions. These images are a valuable newlinedata source for the Land Use - Land Cover (LULC) applications, such as urban planning, newlineenvironmental management, change detection, road map making, town planning, newlineand intelligent transportation system, etc. LULC information is the basis for newlineunderstanding the complex interaction between human activity and changes in the newlineenvironment on a global scale. Now-a-days Deep learning (DL) methods have gained newlineenormous attention in LULC applications, including object detection, semantic segmentation, newlineand classification, in addition to more standard computer vision applications. newlineThey have significantly outperformed state-of-the-art DL methods in a variety newlineof disciplines and have gained a lot of success in the academic and professional newlineworlds. However, in addition to road extraction and change detection, remote sensing newlineimage analysis entails several pre-processing processes and is method-dependent. newlineInitially, we investigated and developed a siamese-based dilated depthwise separable newlineconvolution (DWconv) network shortly called (SDDSCNet) for addressing newlinechange detection problems from VHR satellite images. This siamese network gets newlinetrained by areas of overlap of the input imagery from satellites and transfers the newlineweights in two networks. This network s goal is to use less layers of architecture and newlineminimise computing costs by substituting dilated DWconv for regular convolution newlinein siamese-based CNN-convolution neural networks. Furthermore, we presented a newlinedense dilated DWcov (DDWcov) center subsection to completely expand CNN s exposed newlinerange, collect pertinent features, and guarantee semantic segmentation precision. newlineThe present research uses the UDWT - Undecimated Discrete Wavelet Transform newlinefusion for multidimensional and temporal examination of different resolution newlineinputs to refine the difference map and generate a much deeper information change newlinemap as a post-processing