Remote sensing image binary change detection method based on feature deviation alignment

Abstract

The invention discloses a remote sensing image binary change detection method based on feature deviation alignment. The method comprises the following steps: 1, constructing a dual-time-phase remote sensing image binary change detection data set and carrying out preprocessing; 2, constructing a binary change detection model based on feature deviation alignment, and giving a dual-temporal remote sensing image to obtain a change region prediction result and a change region auxiliary prediction map; 3, respectively calculating a main loss function and an auxiliary loss function by using a real change area label result, a prediction change area result and a change area auxiliary prediction map, carrying out back propagation on a gradient according to loss to update a model, stopping training until a loss value converges, and storing a model structure and a model weight; and 4, predicting test set data by using the model weight trained in the step 3. According to the invention, the false detection phenomenon of the change area of the dual-time-phase remote sensing image caused by factors such as multi-view shooting, too many high-rise buildings or large topographic relief can be effectively solved.

Description

[0001] field of invention [0002] The invention belongs to the field of remote sensing image processing, relates to the field of computer deep learning, and in particular relates to a binary change detection method of remote sensing images based on feature deviation alignment. Background technique [0003] The goal of the binary change detection task is to locate the change area and non-change area of ​​the area given two dual-temporal remote sensing images of the same area. With the development of deep learning technology, methods based on twin convolutional neural networks have achieved higher accuracy than traditional methods in binary change detection tasks, which usually use an encoder-decoder architecture, that is, twin convolutional neural networks with shared weights are first used. The network encoder extracts the low-level features and high-level features of the bitemporal image respectively, and then uses the decoder to obtain the difference features and gradually ...

AI Technical Summary

Problems solved by technology

[0004] However, when the binary change detection method based on Siamese convolutional neural network obtains difference features in high-dimensional feature space, it needs to ensure that the extracted high-dimensional features are aligned in the original bitemporal image as much as possible, otherwise it will be due to The problem of feature deviation appears false detection area

However, due to the existence of registration errors and downsampling layers, there will inevitably be a problem of feature deviation between the high-dimensional features of bitemporal images.

The existing binary change detection methods often directly introduce the semantic segmentation model or make certain improvements on the basis of the semantic segmentation model, without considering the problem of feature deviation

Images