结合解混的高光谱区域生长算法研究及应用

发布时间：2018-06-22 04:57

本文选题：区域生长 + 高光谱遥感　；参考：《大连海事大学》2015年硕士论文

【摘要】：随着海上石油运输业的蓬勃发展,溢油事故的发生愈加频繁,其对海洋环境的污染无疑是毁灭性的。事故发生后,航载高光谱遥感监测可以及时提供油区和油膜厚度信息,为溢油量的估计和风险评估提供了有力依据。然而目前大部分的分割算法对高光谱海面溢油图像的划分效果并不十分理想,因此,对其进一步的研究十分必要。基于现有的区域生长算法,论文结合高光谱技术和溢油图像特点,从以下几个方面展开研究工作。(1)结合溢油图像本身的特点,首先采用非监督端元提取思想自动获取种子点。之后,对高光谱溢油图像进行解混,得到与高光谱图像对应的丰度图,以丰度作为距离尺度,在二维的丰度图上进行区域生长。(2)由于海面上获取的图像受波浪和太阳耀光的影响严重,在溢油图像中产生大量的异常点,论文尝试了三种改进的生长方式,分别为邻域均值生长方式、邻域极值生长方式和抛去邻域内异常点后的均值生长方式。将这三种生长方式与原始生长方式一起,分别与上述两个步骤相结合进行实验分析,并与采用阈值分割方法和聚类分割方法得到的结果进行对比,验证了所提方法的有效性。(3)最后,在区域生长过程中的阈值选择方面,以区域边界点的平均梯度和区域内部的方差作为参数构建模型,实现自动阈值最优化选择,确保得到最佳的分割效果。论文分别利用模拟图像、在美国印第安纳州(Indiana)某农林混合试验场获取的AVIRIS高光谱图像数据和蓬莱19-3C平台溢油遥感图像作为实验数据,进行了实验分析和效果对比。通过实验结果证明,论文中提出的结合解混技术和改进的生长方式的区域生长方法是有效且可行的。该方法减轻了错分和漏分的现象,提高了分割的精度,且降低了运行时间。
[Abstract]:With the booming development of offshore oil transportation, oil spill accidents occur more frequently, and the pollution to marine environment is undoubtedly destructive. After the accident, airborne hyperspectral remote sensing monitoring can provide oil area and oil film thickness information in time, which provides a powerful basis for oil spill estimation and risk assessment. However, most of the current segmentation algorithms are not very ideal for the classification of hyperspectral oil spill images, so it is necessary to further study them. Based on the existing region growth algorithm, combining the hyperspectral technology and the characteristics of oil spill image, this paper starts the research from the following aspects. (1) considering the characteristics of oil spill image itself, the idea of unsupervised end element extraction is used to obtain seed points automatically. After that, the hyperspectral oil spill image is unmixed and the abundance map corresponding to the hyperspectral image is obtained. The abundance is taken as the distance scale. (2) because the images obtained on the sea surface are seriously affected by the waves and solar flares, a large number of abnormal points are produced in the oil spill images. They are neighborhood mean growth mode, neighborhood extremum growth mode and mean growth mode after throwing outliers in the neighborhood. These three growth modes are combined with the original growth mode, and the experimental results are compared with the results obtained by using the threshold segmentation method and the clustering segmentation method. The effectiveness of the proposed method is verified. (3) finally, in the aspect of threshold selection in the process of regional growth, the model is constructed with the average gradient of the regional boundary point and the variance within the region as the parameters to realize the optimal selection of the automatic threshold. Make sure you get the best segmentation results. In this paper, the AVIRIS hyperspectral image data obtained from a mixed agricultural and forestry test site in Indiana, USA, and the oil spill remote sensing image of Penglai 19-3C platform are used as experimental data, and the experimental results are compared. The experimental results show that the proposed method is effective and feasible. This method alleviates the phenomenon of false and missing points, improves the accuracy of segmentation, and reduces the running time.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TE88;TP751

【参考文献】