Multi-objective optimized sar image change detection method based on deep belief network

Abstract

The invention discloses a multi-objective optimized SAR image change detection method based on a deep belief network, which mainly solves the problems of relatively large speckle noise in the prior art, unable to maintain the local information of the SAR image and low classification accuracy, and its technical solution is: For two input SAR images Y 1 and Y 2 Using the method of fluctuation parameter division to generate the original difference map D 1 ; Then the original difference map D 1 Denoise to get the denoised difference map D 2 ; from the original difference map D 1 and denoised difference map D 2 Construct two objective functions, and calculate the solution set with the minimum function value of these two objective functions at the same time, and then obtain multiple binary images Q k ; from the binary image Q k and the trained deep belief network to get the final change detection image R k . The invention reduces speckle noise, retains local image information, improves classification accuracy, and can be applied to remote sensing, medical diagnosis and video monitoring.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a SAR image change detection method, which can be used in remote sensing, medical diagnosis and video monitoring. Background technique [0002] Because synthetic aperture radar (SAR) is not affected by factors such as cloud cover and atmospheric conditions, SAR image technology plays an indispensable and important role in people's daily life. The SAR image change detection technology in the SAR image technology plays a particularly critical role. SAR image change detection technology is to study the changes of two or more SAR images in different periods of the same scene. It has a wide range of application scenarios, including monitoring of natural ecology, assessment and prevention of natural disasters, and obtaining information on landform changes. However, SAR image change detection often encounters difficulties. There are many reasons for these difficulties. The ma...

AI Technical Summary

Problems solved by technology

The difference method is the most primitive processing method. The biggest disadvantage of this method is that the suppression of speckle noise is very poor; the advantage of the ratio method is that the speckle noise is suppressed to a certain extent, but the effect is not very obvious. Its biggest disadvantage is There are many additive noises; the logarithmic ratio method converts additive noise into multiplicative noise. This method obtains non-linear stretching through the difference map after logarithmic transformation. Its advantage is that it enhances the contrast between the change class and the non-change class. , but its disadvantage is that the accuracy of the difference map is not high

[0005] The second is the post-classification comparison method. The key to this method is the extraction of change information in the difference map. Its shortcoming is that there is a classification accumulation error problem, which affects the classification accuracy.

Images