The Earth's surface has changed significantly as a result of human activity on the land expanding agriculture and population. To fulfil the growing demand for fundamental human necessities and wellbeing, it is crucial to have correct information on land use and land cover (LULC) and the best methods of using it. Large geographic regions can be found in sufficient detail in satellite photos, both qualitatively and quantitatively. The most effective methods for detecting together static and dynamic biophysical modules on the Earth's surface, which are regularly introduced for mapping LULC, are satellite depending remote sensing (RS) methods. In order to classify RS images into change/nochange classes, image pre-processing is done in this study, and the information content of the satellite images is assessed. In this work, a change detection method for identifying land cover and water bodies is proposed utilizing a stacked ensemble classifier with mean weight residual neural network (MWResNet) and entropy.
Top1. Introduction
In environmental research, remote sensing imageries are frequently employed to spot changes from anthropogenic or natural sources (Hussain, et al., 2013; Gregoire, et al., 2010). Change detection is “the technique of finding differences in the condition of an object or phenomenon by watching it at different periods.” In general, change detection methods are performed based on a two-step pattern. Feature extraction is the first step that is worked based on input images and is used to extract features, and these features are given as input for the decision step (Du, et al., 2012). In the decision step, features extracted from the previous step are classified into “change” vs. “no-change” for getting the final output (Gregoire, et al., 2010). Some methods used for the decision step change image differencing, image rationing, Principal Component Analysis (PCA), and tasseled cap transformations (Ng, et al., 2013). Classification algorithms use similarity or dissimilarity indices to categorize diverse land coverings according to their spectral characteristics (Chignell, et al., 2015). Remote sensing generally employs three classification techniques: supervised, unsupervised, and hybrid (Roy, et al., 2014). Three steps comprise the supervised approach: training the data, classifying using the trained data, and testing the classification outcomes. Instead, an unsupervised learning algorithm training dataset is not required (Shang, et al., 2019). The hybrid technique is worked based on semi-supervised and unsupervised procedures (Jia, et al., 2014). For the categorization of land use and land cover (LULC) for several applications, researchers have presented a variety of supervised and unsupervised approaches, as well as their combinations (Zalpour, et al., 2020; Samadi, et al., 2019; Acharya & Yang, 2004; Acharya, et al., 2016; Mishra & Pant, 2019).
The main Research Objectives are as follows:
Flora, mountains, hills, swamps, and water bodies cover the planet’s surface (Kondraju, et al., 2014). By analyzing two or more photos of the same location taken at various times, change detection tries to find changes on the earth’s surface (Salah, 2017). It can be used for various purposes, including tracking land use and cover changes, urban planning, stopping deforestation, and recovering forests (Chignell, et al., 2015). Thus, the proposed methodology offers a practical alternative that enhances RS photos’ water body area change detection procedure (Uddin, et al., 2022). The following is a description of the proposed work’s main goals:
- •
To introduce a Weighted Local Fuzzy C Means (WLFCM) Segmentation and Multi-scale Cat Swarm Optimisation (MCSO) approach to enhance the final accuracy of the change detection (Alarood, et al., 2022).
- •
The proposed system performs Multi-scale Segmentation to boost the overall accuracy of detection rate to variation in the spatial (Ullah, et al., 2020).
- •
Texture features, morphological features, and Principal Component Analysis (PCA) features were retrieved from the water body image to improve detection accuracy, shorten system runtime, and collect crucial information (Rani, et al., 2021).
- •
The MWResNet (Mean Weight ResNet) architecture and Entropy Parameter Optimization Deep Belief Network (EPO-DBN) approach are introduced to detection regulate to raise the detection rate of the water body, and it is capable of gathering several patterns of information from the dataset.
- •
MWResNet and EPO-DBN execute stacking ensemble classification to increase overall classification accuracy (Saleh, et al., 2021).