Research On Landslide Area Detection Algorithms Based On Remote Sensing Images

In the aftermath of a landslide,the use of satellite remote sensing technology can play a crucial role in disaster mitigation and relief efforts,as well as in planning and construction activities in affected areas.Optical remote sensing satellites have several advantages,including a short revisit period,abundant data,and high resolution,which make them well-suited for postdisaster landslide information extraction and large-scale mapping.However,the occurrence of landslides is often accompanied by adverse weather conditions,which can significantly hinder the acquisition of post-disaster optical remote sensing data.In contrast,Polarimetric Synthetic Aperture Radar(PolSAR)can provide polarization scattering information in the landslide area in a timely manner,without being affected by weather conditions.However,the spatial resolution of PolSAR images is lower than that of optical remote sensing images,and the availability of PolSAR satellites is currently limited,with insufficient data reserves.In light of the different data characteristics and applicable scenarios of these two types of remote sensing images,this paper proposes targeted algorithms to achieve the goal of timely and accurate detection of landslide areas following a landslide event.The main objectives of this study are to design and implement effective algorithms that can leverage the strengths of both optical and PolSAR remote sensing data to achieve accurate landslide detection and mapping.By doing so,this study aims to contribute to the development of more effective and efficient landslide disaster mitigation and relief strategies,and to support more informed planning and construction decisions in areas prone to landslides.:1.An important application of optical remote sensing images after disasters is the extraction of landslide information and large-scale landslide mapping.However,current methods based on deep learning face challenges such as dealing with multiple data types,complex data processing,and high computing resource consumption.To address these issues,we propose a lightweight landslide area detection method that uses post-disaster single-temporal remote sensing images called Reg-SA-UNet++.This approach employs a lightweight network for feature extraction,which reduces the model parameters and enhances feature acquisition capabilities.The attention mechanism is utilized to improve the model’s focus on the landslide and reduce background interference.Our proposed method achieved an overall accuracy improvement of 1.03% and reduced the number of parameters by10.75 M compared to the suboptimal value on the landslide dataset.In the landslide mapping experiment,we obtained an overall accuracy of 97.09%.2.The use of dual-time phase PolSAR data can quickly and fully obtain the scattering information of landslide areas.In response to the problem of inaccurate detection of landslide areas in complex backgrounds with multiple types of terrain and the need to suppress false alarms,a landslide detection method that uses the Technique for Order Preference by Similarity to An Ideal Solution(TOPSIS)to fuse polarization information is proposed,called PolSAR＿LDM＿TOPSIS.Firstly,TOPSIS is used to synthesize multiple polarization feature parameters to obtain suspected landslide areas.Then,a coherence graph is used to eliminate false alarms in suspected landslide areas,resulting in the detection of landslide areas.The proposed method achieved correct rates of 84.73% and 84.48% on two landslide datasets in Shenzhen and Hokkaido,Japan,respectively,while effectively suppressing false alarms,representing an improvement of approximately 11% and 16% compared to existing methods.3.The amount of pre-disaster optical remote sensing data is sufficient,PolSAR images can be obtained in time under post-disaster bad weather,and the combined use of pre-disaster optical and post-disaster PolSAR images can quickly respond to landslide events without being limited by pre-disaster PolSAR images.Aiming at the problem that the current method requires geoscience data assistance and the detection accuracy needs to be improved,a multi-source remote sensing landslide area detection method relying only on image data-LDM＿ORS&PolSAR is proposed.Firstly,the suspected landslide area is extracted from the PolSAR image after landslide based on automatic threshold segmentation,and the area that is easy to interfere with landslide detection in the optical image is extracted based on objectoriented classification,and then the landslide range is determined by combining the two results.On the two landslide datasets in Guizhou and Hokkaido,Japan,the overall accuracy of the method is 87.80% and 97.83%,respectively,and the accuracy of landslide producers is 71.66%and 71.78%,respectively,which is about 27% and 16% higher than the existing method.