顾及测量不确定性的水体悬浮物浓度遥感定量反演方法

发布时间：2018-02-26 16:42

  本文关键词： 海洋光学 遥感定量反演 测量不确定性 悬浮物浓度 极限学习机 随机抽样一致性 N邻近点抽样一致性　出处：《光学学报》2016年07期 　论文类型：期刊论文

【摘要】：在遥感定量反演的地面同步实测环节中,人为因素、环境变化、条件限制等测量不确定性因素会不可避免地引入数据噪声,致使水体悬浮物浓度反演精度降低。为此,提出一种顾及测量不确定性的水体悬浮物浓度遥感定量反演方法,即自适应抽样一致性极限学习机(ASAC-ELM)算法。该算法结合了极限学习机(ELM)、随机抽样一致性(RANSAC)和N邻近点抽样一致性(NAPSAC)方法的优势与特点,利用参数维度自适应地选取RANSAC或NAPSAC算法进行参数估计,避免了ELM算法易受非零均值正态分布数据噪声影响的缺陷。ASAC-ELM算法通过选取局内点(非噪声点)数据建立模型,可去除噪声数据的干扰,提升模型的精度与适应性。通过模拟多组不同数量级且服从非零均值正态分布的随机数,将加性噪声引入训练数据中,实现不同噪声比条件下对ASAC-ELM算法的检验,并与ELM算法、传统反向传播(BP)神经网络算法进行了对比。结果表明,不同噪声比条件下,ASAC-ELM算法的水质悬浮物浓度反演精度高于ELM算法和传统BP神经网络算法,且反演结果稳定性较高。
[Abstract]:In the synchronous ground measurement of remote sensing quantitative inversion, the data noise will inevitably be introduced into the uncertainty factors such as human factors, environmental changes, conditions and so on, which will reduce the accuracy of the inversion of the concentration of suspended matter in the water body. In this paper, a quantitative inversion method for the concentration of suspended matter in water body by remote sensing is proposed, which takes into account the uncertainty of measurement. This algorithm combines the advantages and characteristics of the extreme learning machine (LLM), random sampling consistency (RANSAC) and the N neighborhood sampling consistency (NAPSAC) method, which is based on the adaptive sampling consistency limit learning machine (ASAC-ELM) algorithm, which combines the advantages and characteristics of the extreme learning machine (LLM) and the random sampling consistency (RANSAC). The parameter dimension is used to adaptively select RANSAC or NAPSAC algorithm for parameter estimation, which avoids the defect of ELM algorithm which is vulnerable to the non-zero mean normal distribution data noise. ASAC-ELM algorithm builds a model by selecting local points (non-noise points) data. The interference of noise data can be removed, and the accuracy and adaptability of the model can be improved. By simulating multiple random numbers of different orders of magnitude and applying normal distribution from non-zero mean, additive noise is introduced into the training data. The ASAC-ELM algorithm is tested under different noise ratio, and compared with the ELM algorithm and the traditional BP neural network algorithm. The results show that, The accuracy of ASAC-ELM algorithm under different noise ratio is higher than that of ELM algorithm and traditional BP neural network algorithm, and the stability of the inversion results is higher than that of the traditional BP neural network algorithm.
【作者单位】： 中国地质大学(武汉)信息工程学院;
【基金】：国家自然科学基金(41501459;41301380)
【分类号】：X87
，

本文编号：1538828