适用于洪水遥感SAR图像的CV分割模型研究

A Study on CV Segmentation Model for Flood SAR Images

摘要: 洪水给人们生活造成很大的不便。为了尽快恢复生产生活，及时获取洪水受灾区域的范围及受灾区域发生的变化，对于灾情分析和灾后重建至关重要。由于受到相干斑噪声和洪水受灾区域地形起伏较大的因素的影响，合成孔径雷达(SAR)探测到的洪水遥感图像存在噪点较多且灰度分布不均匀的缺点，成像效果不佳，因此很难对其实现准确地图像分割。为了解决这个问题，本文提出一种基于水平集方法的改进CV分割模型，以实现快速准确地分割洪水遥感图像。CV分割模型是一种在医学图像分割领域很常见的基于区域和边缘信息的图像分割模型，能够得到较为精确的分割曲线，但是存在对噪声敏感、过度依赖区域灰度均值的缺陷。对于存在大量的噪声且图像灰度分布不均匀的洪水遥感图像，CV分割模型效果不佳。为了减少噪声在分割曲线演化过程中的干扰，本文在CV分割模型的能量函数中加入一个整体控制项，以减少噪声的影响。此外，为了提升CV分割模型对灰度不均匀的洪水遥感图像的分割能力，对处于不同区域的像素点采用自适应动态权重的方法进行分割曲线拟合。实验结果表明，与传统的CV分割模型和其他的图像分割模型相比，本文提出的改进CV分割模型在灰度分布不均性较强的洪水遥感图像上有更出色的分割精度，证明了本文提出的改进CV分割模型的有效性。

Abstract: The flood caused great inconvenience to people’s life. In order to resume production and living as soon as possible, obtaining the scope of affected areas and the changes in affected areas in time is very important for disaster analysis and post-disaster reconstruction. Due to the influence of speckle noise and large topographic relief in affected area, the flood remote sensing images detected by Synthetic Aperture Radar (SAR) have the disadvantages of many noise points and uneven gray distribution, and the imaging effect is not good. Therefore, it is difficult to achieve accurate image segmentation results. To solve this problem, in this paper, an improved CV segmentation model based on level set method is proposed to achieve rapid and accurate segmentation of flood remote sensing images. CV segmentation model based on region and edge information is very common in the field of medical image segmentation. It can obtain a more accurate segmentation curve, however, it has some defects such as sensitive to noise and over-relying on regional gray mean. For flood remote sensing images with large amount of noise and uneven gray distribution, CV segmentation model is not effective. In order to reduce the interference of noise in the process of forming the segmentation curve, an integral control is added into the energy function of CV segmentation model in this paper to reduce the influence of noise. In addition, in order to improve the segmentation ability of CV segmentation model on the flood remote sensing images with uneven gray scale, the adaptive dynamic weight method was used to fit the segmentation curve for the pixel points in different regions. The experimental results show that, compared with the traditional CV segmentation model and other image segmentation model, the improved CV segmentation model proposed in this paper has better segmentation accuracy on the flood remote sensing images with strong uneven gray distribution, which proves the effectiveness of the improved CV segmentation model proposed in this paper.