最优分割尺度支持下高分遥感影像水土资源信息分类

Soil and Water Resources Information Classification in High Resolution Images with Optimal Segmentation Scale

为提升水土资源信息分类精度,以无人机航拍获取的高分辨率影像为实验对象,提出了最优分割尺度和决策树支持下的对象级影像分类方法。首先,根据影像内部的同质性和异质性,建立了分割质量函数,通过该函数获取了最优分割尺度;然后,提出了基于光谱信息和面积信息的最优分割尺度评价模型对分割结果进行评价;最后,引入决策树规则机制,完成了水土资源信息分类,并与最大似然法分类结果进行对比。研究结果表明:所建立的分割质量函数能准确获取最优分割尺度,有效避免了人工分割带来的主观性,所提方法分类总体精度为86.78%,最大似然分类方法总体精度为77.59%,在分类精度上有较大提升。

With the rapid development of agricultural informationization, the demand for accuracy and reality of regional soil and water resources information data becomes higher and higher. The progress of remote sensing technology makes the selectable data source richer. High spatial resolution images contain rich shape and texture information which are widely used in soil and water resources survey, while traditional image classification method cannot satisfy the requirement any more. Because of this, unmanned aerial vehicle （UAV） images were used as experimental objects, and the image objectoriented classification method based on optimal segmentation scale and decision tree was proposed. Firstly, a segmentation quality function was established based on internal homogeneity and heterogeneity of images, and the optimal segmentation scale was obtained according to this function. Then, optimal segmentation scale evaluation model based on spectral and area information was proposed to evaluate segmentation result. Lastly, soil and water resource information classification was completed by introducing decision tree rule mechanism, and compared with the maximum likelihood classification results. The experimental results showed that the segmentation quality function can obtain optimal segmentation scale accurately, and avoid the subjectivity of manual segmentation. The overall accuracy is 86.78% and compared with 77.59% of maximum likelihood classification method has a great improvement in classification accuracy.