SAR image road extraction method and device based on semantic segmentation and conditional random field

Abstract

The invention relates to an SAR image road extraction method and device based on semantic segmentation and a conditional random field. The method comprises the following steps: acquiring an SAR road image sample, inputting the SAR road image sample into a preset semantic segmentation model, carrying out feature extraction on the SAR road image sample through a multilayer convolutional neural network, inputting extracted shallow features into a parallel channel of a decoder, inputting the deep features obtained through extraction into a spatial pyramid module to be processed, obtaining the encoder features, inputting the encoder features into a decoder to be subjected to up-sampling and inputting the features into a parallel channel, and outputting a road extraction prediction result through multiple times of up-sampling; then introducing a second-order point pair full-connection conditional random field, and outputting an SAR image road. By adopting the method, the accuracy and comprehensiveness of road extraction can be improved.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a method and device for extracting roads from SAR images based on semantic segmentation and conditional random fields. Background technique [0002] Synthetic Aperture Radar (SAR) is an active imaging radar with multiple polarization modes and imaging conditions not affected by weather conditions. The single-polarization SAR extraction algorithm is based on the nature of specular scattering of radar waves by smooth roads, and mainly uses gray features to extract roads. The current method can be divided into threshold segmentation and region growing according to traditional image processing methods. With the deep learning in the field of computer vision With the rise of remote sensing data and the abundance of remote sensing data, semantic segmentation models have achieved significant performance improvements in SAR image pixel-level classification tasks such as sea an...

AI Technical Summary

Problems solved by technology

[0003] However, in Mode 1, prediction is made on image slices to avoid the loss of information caused by downsampling, but the network cannot extract semantic information outside the slices, resulting in a greatly reduced feature receptive field. Due to the complex structure of SAR airport images and The surrounding interference is serious. This mode is prone to regional judgment errors, resulting in a decrease in the segmentation effect instead of an increase. In addition, this mode takes a long time and has little application value.

In mode 2, since the image is first down-sampled, and the feature size is reduced in order to extract high-dimensional features during the forward pass of the network, the feature map obtained by this method can obtain a large receptive field feature, but the image details are seriously lost.

Images