高光谱遥感影像除了包含普通2维影像所具有的空间信息还包含了1维光谱信息,传统的针对2维影像的分割方法不能很好地应用于高光谱遥感影像。为此,本文提出一种能够同时处理多波段影像的高光谱遥感影像矢量C-V模型分割方法。首先选出高光谱遥感影像中目标与背景对比度较大的波段,并通过计算波段相关系数,去除其中的冗余信息形成新的波段组合,进而根据所确定的波段组合构建高光谱遥感影像矢量矩阵;在此基础上,构造基于该矢量矩阵的矢量C-V分割模型。模型中通过引入基于梯度的边缘引导函数,在保留传统C-V模型基于区域信息进行影像分割的基础上,利用影像的边缘细节信息,增强了模型在异质区域和复杂背景情况下对目标边缘的捕捉能力,提高了对高光谱遥感影像的分割精度和速度。最后利用HYPERION数据进行仿真实验,并将实验结果和传统C-V模型和相关方法进行了对比,结果表明,本文方法能够在短时间内有效地分割高光谱遥感影像,与传统方法相比,具有分割精度更高运算速度更快的特点。

This study on the vector C-V model and hyperspectral remote sensing image aims is to segment a hyperspectral remote sensing image. A hyperspectral remote sensing image contains not only general two-dimensional image spatial information but also have one-dimensional spectrum information. Thus, traditional methods of two-dimensional image segmentation are unsuitable for hyperspectral remote sensing images. To solve this problem, we propose a hypespectral remote sensing image vector C-V model segmentation method based on band selection, which can deal with the multiband images at the same time. Method First, bands of goals and backgrounds contrast that exhibit a significant contrast were chosen based on the band correlation coefficient. Then, the greater relevance band-by-band correlation coefficient was removed, and a new band combination was formed. Finally, a hyperspectral remote sensing image vector matrix was built. On these basis, we can construct a vector C-V model that takes full advantage of this vector matrix while introducing a gradient-based edge guide function. Result Numerical experiments were conducted on HYPERION data, and these experiments were compared with the traditional C-V model and Wang & Jin method. The result shows that the proposed method can immediately segment a hyperspectal remote sensing image effectively, and it not only has a lower fase-positive ratio and false-negative ratio but also a smaller error ratio. These results prove that the segmentation of the proposed model is more effective than that of the traditional C-V model and Wang & Jin method. In sum, compared with the traditional C-V model and Wang & Jin method, the proposed model improved the segmentation speed and accuracy. Conclusion The proposed model does not retain the traits of conventional C-V model, which is based on the regional information. Rather, it increased the ability of capture the boundary of the target in heterogeneous regions and complex background by using image edge details. However, the method has some shortcomings. For instance, it uses only the gray level information without the spectral information of hyperstpectral remote image during the process of segmentation, which causes a small amount of error in its results. Using the combination of spatial information and spectral information effectively in the process of the segmentation is a problem that needs further research.