Change detection in Landsat 8 imagery using object-based image analysis with particle swarm optimisation

This paper addresses the problem of classification of hyperspectral remote sensing image and detection of any changes in the land use pattern using it. Traditional pixel-based classification approaches often fail to achieve acceptable accuracy in classifying Landsat images because of its complexity. Therefore, the present work aims to apply object-based image analysis (OBIA), which is a concept that combines segmentation and classification together into one unit. We propose a hybrid OBIA architecture, augmented with particle swarm optimisation (PSO) technique to fine tune different hyperparameters involved with it. We present its success on a classification problem where two sets of Landsat-8 images captured in the year 2016 and 2017 are considered as input and OBIA is applied for classifying the images into four major land use classes and detect any changes in these classes over a period of time. The results are then compared with best pixel-based classification approach known as random forest (RF) classifier to determine its effectiveness in classification of hyperspectral images. Statistical measures like precision, overall accuracy and kappa coefficient are used as a parameter for comparison.