Enhanced Change Detection Analysis of Urban Land Use and Land Cover in Vijayawada City: Integrating Artificial Neural Networks and Mahalanobis Distance Classification

The main goal of the study is to find out the changes that are occurring on the land due to the change of surface cover by its use during the period 2001 to 2020 for Vijayawada city, Andhra Pradesh. This is found by doing digital image processing using two different classifiers Artificial Neural Networks (ANN) and Mahalanobis-Based Distance (MBD)-based novel supervised classification and comparing both to find which is more accurate. For digital image processing, satellite images downloaded for image classification from the United States Geological Survey (USGS) are used as the samples. The samples are downloaded for three different years 2001, 2011, and 2020 consisting of the urban study region. Images were acquired from both Landsat 7 ETM+ and Landsat 8. Two groups of classifiers and three samples for each group totaling to six samples were used to test the accuracy. With pre-test power at 80%, alpha at 0.05 and CI at 95%, a statistical examination was done. A p value of 0.13 denotes that there is no significant difference between the groups. The percentage of broadly classified six regions are found by doing novel supervised classification by both the algorithms and noted down. The analysis is done for the key outputs overall accuracy (OA) and kappa coefficient (KC). The obtained OA is 97.10 ± 1.61 as mean and SD for ANN, 92.49 ± 6.76 as mean and standard deviation for MBD. For KC, 0.93 ± 0.05 is derived as mean and standard deviation for ANN, 0.8574 ± 0.1539 as mean and SD for MBD classification, respectively. Artificial Neural Networks are the best approach to find the land cover changes using the satellite images compared to the Mahalanobis distance-based classification from the results of the research.