MOBS-TD: Multiobjective Band Selection With Ideal Solution Optimization Strategy for Hyperspectral Target Detection

Band selection (BS) is a crucial concept within the realm of remote sensing, involving the selection of the most suitable bands to accurately capture features of landforms and surfaces. Despite the promising results achieved by many existing methods, certain limitations remain. First, most methods r...

Full description

Saved in:
Bibliographic Details
Published in:IEEE journal of selected topics in applied earth observations and remote sensing 2024, Vol.17, p.10032-10050
Main Authors: Sun, Xudong, Lin, Peng, Shang, Xiaodi, Pang, Huijuan, Fu, Xianping
Format: Article
Language:English
Subjects:
Band selection (BS)
Competitiveness
Evaluation
False alarms
Feature extraction
hyperspectral imagery
Hyperspectral imaging
Landforms
multiobjective optimization (MO)
Multiple objective analysis
Object detection
Optimization
Pareto optimization
Redundancy
Remote sensing
Sociology
Target detection
Task analysis
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Band selection (BS) is a crucial concept within the realm of remote sensing, involving the selection of the most suitable bands to accurately capture features of landforms and surfaces. Despite the promising results achieved by many existing methods, certain limitations remain. First, most methods rely on a single criterion for band evaluation, leading to an incomplete assessment and limited generalizability of bands. Second, there is a lack of emphasis on target detection; thus, some BS techniques commonly used for classification are less effective for detection. Therefore, this article proposes MOBS-TD, a multiobjective optimization (MO) based BS method specifically designed for target detection, which aims to select bands with better target separation and stronger robustness across various application scenes. Initially, we develop an MO model with three objectives and introduce a novel metric to quantify the target-background separability of bands. Subsequently, a weighted similarity to ideal solution strategy is developed to clearly describe the dominance relations and strike a balance among multiple objectives in evolution. In addition, we devise an evaluation mechanism based on the ratio of maximum to submaximum, which is devised for selecting the optimal solution from the Pareto front, which has been empirically validated to be effective in reducing false alarms. Extensive experiments on real-world datasets demonstrate the competitiveness of MOBS-TD in remote sensing applications.
ISSN:1939-1404
2151-1535
DOI:10.1109/JSTARS.2024.3402381