An Effective Deep Learning Approaches on Change Detection using Multi Spectral Satellite Images

Abstract

The wealth of the country depends on the natural wealth hence preserving the na- newlineture is important. Remote sensing technology is helpful to track the status of nature. newlineChange Detection (CD) is a process commonly used in remote sensing to monitoring and understanding environmental changes. CD refers to the process of identifying differences or variations between two or more sets of images captured at different newlinetimes. It is commonly used in various and#57344;elds such as remote sensing, computer vision, newlinesurveillance systems, and data analysis. CD techniques aim to highlight the areas or newlineaspects that have undergone signiand#57344;cant changes, allowing users to better understand newlineand interpret the variations. CD is a critical task for monitoring deforestation, forest degradation, reforestation, and other land cover changes. Techniques such as image newlinedifferencing, thresholding, and spectral indices, Normalized Difference Vegetation In- newlinedex (NDVI) are applied to identify areas of deforestation, regrowth, or other changes newlinein vegetation. For CD, many techniques were performed. The main objective of this newlinethesis is to developed a CD method which combine various model along with opti- newlinemization strategy for multi-spectral satellite image. Here, different techniques are proposed for detection of changes on earth s surface. At and#57344;rst, a novel hybrid optimization based CD approach. Initially, the multi-spectral satellite image is pre-processed to remove artefacts from the image. Then the essential features are extracted through seven vegetation indexes NDVI, Simple Ratio (SR), Soil-Adjusted Vegetation Index (SAVI),Weighted Difference Vegetation Index (WDVI), Global Environmental Monitoring In- newlinedex (GEMI), Modiand#57344;ed Soil-Adjusted Vegetation Index (MSAVI), and Modiand#57344;ed Simple newlineRatio (MSR) for segmenting the image. The image segmentation is carried out using the newlineUNET, which is tuned using the proposed hybrid optimization, namely, the Wader hunt newlineoptimization (WaHO) algorithm, to obtain a more accurate detection. The WaHO is newlinedesigned by integratin