A SAR Image Remote Sensing Detection Method Based on Joint Spatial Fully Connected CRF

To address the localization issues caused by the loss of in-situ information in traditional dual phase pictures change detection methods, this paper proposes a SAR image change detection method based on a fully connected Conditional Random Field (CRF) within a geo-difference joint space. Firstly, a...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhang, Jianlong, Luo, Jiangtao, Xiong, Huihuang, Zhang, Leichang, Chen, Chen, Li, Wan, Yang, Fei, Li, Huimin, Gan, Qaun
Format: Conference Proceeding
Language:English
Subjects:
Change Detection
Conditional random fields
Convergence
FC-CRF
Iterative methods
Location awareness
Neural networks
Radar polarimetry
Remote sensing
SAR Image
Self-supervised learning
Semantics
Siamese Neural Network
Simulation
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:To address the localization issues caused by the loss of in-situ information in traditional dual phase pictures change detection methods, this paper proposes a SAR image change detection method based on a fully connected Conditional Random Field (CRF) within a geo-difference joint space. Firstly, a multi-scale fusion Siamese neural network is designed to initially locate change regions in remote sensing images based on their geographic spatial information, serving as the unary potential function of the CRF model. Secondly, a differential image is used to construct a pairwise potential function based on grayscale and positional information. Finally, the final change localization is achieved through the iterative convergence of the energy function. Simulation results indicate that the proposed method yields change detection results with complete regions and accurate boundary localization, demonstrating competitive detection performance compared to other state-of-the-art methods.
ISSN:2770-2677
DOI:10.1109/SmartIoT62235.2024.00045