Dimension Reduction for Hyperspectral Remote Sensor Data Based on Multi-Objective Particle Swarm Optimization Algorithm and Game Theory

Abstract

Information entropy and interclass separability are adopted as the evaluation criteria of dimension reduction for hyperspectral remote sensor data. However, it is rather single-faceted to simply use either information entropy or interclass separability as evaluation criteria, and will lead to a single-target problem. In this case, the chosen optimal band combination may be unfavorable for the improvement of follow-up classification accuracy. Thus, in this work, inter-band correlation is considered as the premise, and information entropy and interclass separability are synthesized as the evaluation criterion of dimension reduction. The multi-objective particle swarm optimization algorithm is easy to implement and characterized by rapid convergence. It is adopted to search for the optimal band combination. In addition, game theory is also introduced to dimension reduction to coordinate potential conflicts when both information entropy and interclass separability are used to search for the optimal band combination. Experimental results reveal that compared with the dimensionality reduction method, which only uses information entropy or Bhattacharyya distance as the evaluation criterion, and the method combining multiple criterions into one by weighting, the proposed method achieves global optimum more easily, and then obtains a better band combination and possess higher classification accuracy.

Computational Intelligence in Remote Sensing: An Editorial

“…Band selection is often based on information entropy or interclass separability as evaluation criteria. In [7], inter-band correlation is considered as the guide for band aggregation, and information entropy and interclass separability are the dimension reduction evaluation criteria. They apply a multi-objective particle swarm optimization (PSO) algorithm because of its easy implementation and rapid convergence.…”

“…The appearance of new and more powerful remote sensing technologies has produced a surge of remote sensing data to be processed for a variety of applications in the natural sciences, such as agriculture, forestry or ecological monitoring, and others, such as the automotive industry. This special issue contains a broad sample of such applications, including indoor crowd detection and localization by means of anonymous and non invasive sensors [1]; applications in the automotive industry, such as the detection of the incoming obstacles/vehicles by on-board radar [2], and the detection of highly contaminant vehicles [3]; land cover segmentation for several purposes, such as ecological monitoring [4,5], and land uses [6,7].…”

“…The surging abundance of data has been accompanied by a blooming of computational resources and methods for analysis, such as evolutionary and random search algorithms, particle swarm optimization (PSO) [7], and ant colony optimization (ACO) [8], which are often extended as multi-objective optimization algorithms. Computational intelligence tools are increasingly being used for pre-processing, enhancement, classification, and construction of thematic maps, change detection, target detection, sub-pixel resolution analysis, and other general processes.…”

Analysis of the Key Factors Affecting the Capability and Optimization for Magnetostrictive Iron-Gallium Alloy Ambient Vibration Harvesters

“…However, the billions of batteries produced each year are unsustainable, their disposal poses environmental problems, and their limited service life poses a challenge to the long-term/autonomous operation of the devices. The rapidly growing market for solid-state electronics is leading the development of ultra-low-power devices and self-powered devices [1], such as wearable electronic devices [2], biosensors [3], IoT and remote sensors [4,5]. Accordingly, the world is very interested in reducing carbon dioxide emissions and developing more renewable energy technologies to generate electricity.…”

Composite Clustering Sampling Strategy for Multiscale Spectral-Spatial Classification of Hyperspectral Images

“…Hyperspectral image (HSI) is acquired by dedicated hyperspectral cameras, which contains the spectral information of a same ground object in hundreds of continuous bands [1]. Compared with traditional remote sensing images such as RGB three-band remote sensing images and multispectral remote sensing images, the imaging bands of HSIs have greatly increased.…”