SAR image water probability drawing method in combination of prior probability estimation

Abstract

The invention discloses an SAR image water probability drawing method in combination of prior probability estimation. The method comprises the following steps of 1, establishing an SAR image pixel backward scattering coefficient distribution statistics model hypothesis; 2, estimating a water distribution prior probability; 3, estimating a distribution parameter according to a researched area imagebackward scattering coefficient sigma0; and 4, calculating a condition probability that the pixel belongs to the water. According to the method, Bayesian inference is utilized for performing Gauss distribution hypothesis on the researched area image pixel backward scattering coefficient, classifying the pixels to a water kind and a no-water kind according to a k-means clustering algorithm, calculating the researched area water pixel proportion as the estimated value of the water distribution prior probability, and finally according to the estimated of the water distribution prior probability,overlapping the backward scattering coefficient theoretical probability density function on a statistics distribution histogram, finishing model parameter estimation by means of nonlinear least square fitting, and obtaining the probability that each pixel of the researched area image belongs to water.

Description

technical field [0001] The invention relates to water body remote sensing extraction technology, in particular to a SAR image water body probability mapping method combined with prior probability estimation. Background technique [0002] Water body extraction from remote sensing images is a typical application of remote sensing in the water conservancy industry. SAR (synthetic aperture radar) data is suitable for the extraction and monitoring of flood disaster information due to its all-weather and all-time imaging characteristics, and plays an important role in the field of water conservancy remote sensing. The process of water body extraction can be attributed to the process of classifying ground objects and segmenting images. The classic remote sensing image water body extraction methods include threshold segmentation, classifier segmentation combining multiple features, segmentation based on level set theory, image segmentation based on energy function, etc. When the e...

AI Technical Summary

Problems solved by technology

However, in the case of not relying on other auxiliary information other than images, the existing probabilistic mapping methods simply choose to set the prior probability to the default value of 0.5, which will make the statistical model of the estimated ground object backscatter coefficient insufficient Accurate, reducing the precision of probability mapping results

Images