An image change detection method based on self-organizing map and deep neural network

Abstract

The invention relates to an image change detection method based on a self-organizing map and a deep neural network, comprising the following steps: 1. Using a mean ratio operator MMEAN based on a median filtering process to process two SAR phase images to generate a difference image ; 2. Use the self-organizing map network SOM to perform unsupervised clustering on the difference map, divide the pixels in the difference map into no change class, noise class and change class, and obtain a preliminary change map; 3. Use an automatic noise reduction encoder Construct a deep neural network DNN, use noisy pixels as a training set to train the DNN network, so that it can recognize the characteristics of the noise area; 4. Input the changed pixels into the trained DNN network, judge and remove the changed pixels The residual noise pixels in the points, so as to obtain the final change map based on the preliminary change map. This method is beneficial to reduce the missed detection rate and false alarm rate of synthetic aperture radar image change detection.

Description

technical field [0001] The invention relates to the technical field of image change detection, in particular to an image change detection method based on a self-organizing map and a deep neural network. Background technique [0002] Synthetic Aperture Radar (SAR) surveillance images, as a kind of radar imaging, can obtain high resolution under extreme meteorological conditions compared with optical surveillance images. SAR images are widely used, such as environmental protection, agricultural survey, urban research, forest resource monitoring, etc. The change detection technology based on SAR images can detect the changes of physical targets between observation images for SAR images of the same geographical range and at different times. From the previous literature, SAR image change detection technology has shown great application potential. However, the background information of SAR image is complex, and the texture is often not obvious, which is covered by speckle effect...

AI Technical Summary

Problems solved by technology

However, the traditional ratio operator cannot reasonably balance the suppression of noise and the preservation of edge regions

2. In terms of difference map analysis, traditional analysis methods are often limited by the generation process of the difference map

However, there are usually false alarm points (unchanged pixels with changing characteristics, often referred to as noise) in the change map. At present, there is still a lack of further processing of false alarm points in the change map at home and abroad.

Images