Intelligent Approach for Analysing and Forecasting Land Changes using Multispectral Images

Abstract

Monitoring the changes in land over time is useful to understand the impacts of human activities in the environment. Urban change detection using satellite images plays a major role in research on global environmental change identification and management of natural resources. With availability of multi-temporal satellite data, the proposed work aims to analyse the change on land in Indian regions. This work groups the regions of land with a Firefly Algorithm based clustering approach, which optimizes the cluster center identification process when compared to a conventional clustering approach, as a process of analysing changes with available multi-temporal data. Based on the vegetation, water, and built-up index values, clusters are labelled into appropriate land regions. The extent of change is then assessed using spatio-temporal information available and this helps to identify the pattern of change. Moreover, a deep learning technique, namely Regression-Long Short Term Memory (LSTM) network, is used to forecast future changes, which may be useful in managing urban resources. Understanding how the land has changed in the past, present, and predicted changes in the future may help in making right decisions. Experiments are carried out on Landsat 8 data from the Indian region to identify changes in land using unsupervised learning techniques. A Silhouette index of 0.88 was obtained on average by K-means clustering with three clusters, while 0.93 was obtained using Firefly integrated K-means clustering. Future land image forecasts generated by LSTM are compared to the actual image using the Structural Similarity Index (SSIM) and Root Mean Square Error (RMSE), resulting in SSIM of 0.85 and RMSE of 0.41 when the forecasted spectral band images are stacked into a multispectral image, indicating the effectiveness of the forecasting approach.