耦合最大熵模型的离散粒子群水体提取

Coupling Discrete Particle Swarm Optimization Algorithm with Maximum Entropy Model for Water Extraction from Remote Sensing Image

基于光谱匹配的离散粒子群(SMDPSO)是一种高精度的水体提取方法,但是,由于该方法过度依赖于光谱的相似程度,易将与水体具有相似光谱的地物误判为水体.利用最大熵模型替换光谱匹配方法,提出了耦合最大熵模型的离散粒子群方法(MEDPSO).对全球6个典型区域进行了水体提取,结果表明,在提取包含与水体具有相似光谱的雪冰、云及云阴影等Landsat 8_OLI影像水体时,SMDPSO方法误判率高,水体提取的总体精度和Kappa系数低.MEDPSO方法则能显著抑制雪冰、云及云阴影等因素影响,极大地提高了总体精度和Kappa系数.与SMDPSO方法相比,MEDPSO方法极大地降低了相似光谱对水体提取的影响,能在多种情况下精确地提取水体,扩展了离散粒子群在提取水体方面的应用.

Spectral matching based discrete particle swarm(SMDPSO)is a high-precision method for extracting water bodies.However,because this method relies too much on the degree of similarity of spectra,it is easy to misidentify features with similar spectra to water bodies.In this paper,the maximum entropy model is used to replace the spectral matching method,and a discrete particle swarm method coupled with the maximum entropy model(MEDPSO)is proposed.Water extraction was performed on six typical regions around the world.The results show that the SMDPSO method has a high false positive rate when extracting water bodies containing Landsat 8_OLI images such as snow,ice,clouds,and cloud shadows with similar spectra to the water body.Kappa coefficient is low.The MEDPSO method can significantly suppress the influence of factors such as snow and ice,clouds,and cloud shadows,and greatly improves the overall accuracy and Kappa coefficient.Compared with the SMDPSO method,the MEDPSO method greatly reduces the effect of similar spectra on water extraction,can accurately extract water in many cases,and will expand the application of discrete particle swarms in water extraction.