Very high-resolution satellite image segmentation using variable-length multi-objective genetic clustering for multi-class change detection

Abstract

The baseline approaches on satellite image segmentation problem suffer from issues like sensitivity towards initialization, local optima solutions, a predefined number of output clusters, single-objective optimization, etc. To solve these challenges, this study proposes a unique pixel-level Multi-Spectral (MS) very high resolution (VHR) image segmentation algorithm based on variable-length multi-objective genetic clustering. We propose a new approach to update solutions by retaining variable length property throughout the optimization process. The resulting clustering algorithm contains a set of near-Pareto-optimal solutions. A map that has a scale of less than 1/10000 is called a large-scale map. We propose a large-scale change detection technique as an application of the proposed image segmentation algorithm. Solving Land-use/Land-Cover (LULC) change detection problems in a congested area is a complex task. This study considers the dataset from Pleiades-HR 1B, and Landsat 5 TM sensors in the experimental study. An extensive quantitative and qualitative analysis is performed to validate the superior performance of the proposed method with different state-of-the-art techniques.

Keywords

• Remote Sensing
• VHR Image
• LULC Segmentation
• Variable-length NSGA-II
• LULC Change Detection
• Modified Crowding Distance