Characterisation of Landuse / Landcover Changes and its Comparison in Vijayawada City using Artificial Neural Networks with Minimum Distance to Mean
K. Pavan Venkat, Vidhya Lakshmi Sivakumar
Abstract
Aim: The main objective of this paper is to find the change detection of a Vijayawada city, Andhra Pradesh. Digital image processing is done by performing image classification, the regions separated into 6 different major classes based on its use and by the cover it is found for a period of 20 years from 2001 to 2020 and comparing the algorithms Artificial Neural Networks (ANN) with Minimum Distance to Mean (MDM) classification to predict which gives higher accuracy. Materials and Methods: The images of the city are collected from the satellites landsat 7ETM+ (Enhanced Thematic Mapper plus) &landsat 8 for novel supervised classification from the United States Geological Survey (USGS) earth explorer. Those images are preprocessed for image classification and used for novel supervised classification with a total sample size of N= 6 with 3 samples from 2 groups with different years. The pretest power is to be determined with 80% with an alpha value of 0.05 and CI of 95%. Results and discussion: Based on the results from the SPSS (statistical analysis software) we got p=0.04 for overall accuracy which is p<0.05 for single tailed test, it is evident that there is significant difference between the two groups by the independent sample t test. The novel supervised classification is done for the samples and we have achieved the percentage of land cover, those are taken and we also got key outputs as Overall accuracy and kappa coefficient. We have achieved 97.1067±1.61593 as mean and SD for ANN, 88.8733±6.68356 as mean and SD for MDM for overall accuracy. The kappa coefficients are as 0.9399±0.0550 as mean and SD for ANN, 0.7757±0.1447 as mean and SD for MDM. Conclusion: According to the current research findings, even though there is no significant difference between groups we can say that Artificial Neural Networks provide more accurate classified output and we can say that ANN is better among those two algorithms.