Coupled translation network-based multi-modality image change detection method

Abstract

The invention discloses a coupled translation network-based multi-modality image change detection method and mainly solves the problems that an existing multi-modality image change detection method islow in precision and low in robustness. The method comprises the following implementation steps: (1) setting structures and parameters of two translation networks; (2) inputting two multi-modality images, and calculating a Jason-Shannon divergence distance between the two images and a pixel-not-changing probability coefficient; (3) training the first translation network to obtain a translation result pattern of the first image; (4) training the second translation network to obtain a translation result pattern of the second image; (5) updating the pixel-not-changing probability coefficient according to the two translation results; (6) repeating the steps from (3) to (5) in sequence till a network target function value is stabilized; (7) obtaining a difference pattern according to the two translation results; (8) clustering the difference pattern to obtain a final change detection pattern. The coupled translation network-based multi-modality image change detection method has the advantages of accurate detection and high robustness and can be applied to image translation, model identification and target tracking.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a heterogeneous image change detection method, which can be used for image generation, pattern recognition or target tracking. Background technique [0002] Change detection is a technique for detecting changes in an area by analyzing a set of images taken at different times in the same location. According to different image sources, change detection can be divided into homologous image change detection and heterogeneous image change detection. Among them, homologous images refer to images taken by the same sensor, which have the same attributes, and the pixels in the unchanged area are linearly correlated, so that the difference between pixels can be directly compared to obtain a difference map; heterogeneous images are images obtained by different sensors, For example, synthetic aperture radar SAR images and optical images, their different statistical properties betwe...

AI Technical Summary

Problems solved by technology

However, this method only calculates the unchanged area and uses a fixed parameter to judge whether the pixel has changed or not, resulting in low accuracy when dealing with images with many changed areas or multiple targets.

[0004] Since the above-mentioned classification-based heterogeneous image change detection methods are not accurate and require human intervention, the unsupervised convolutional neural network-based method has a small application range

Images