Two-Phase Architecture For Detection Of Landslides Via Satellite Images

Landslides are severe geological phenomena that often result in the loss of human lives, destruction of property, and interruption of economic activities. The use of image-based techniques in landslide investigations plays a pivotal role in the assessment of vulnerability to landslides and risk. Satellite imagery has been extensively used in practical applications for conducting such studies; yet, it requires substantial allocation of labor and time constraints. This paper presents a novel approach for the detection and segmentation of landslide zones using satellite pictures. The proposed framework is built on a two-phase data-driven methodology that utilizes image analysis techniques. During the first step, the Faster-RCNN technique is used to train an object identification model to identify the precise position of landslides within satellite pictures on a large scale. The suggested and displayed boundary boxes depict the locations of each landslide. The second stage involves the partitioning of the satellite photographs into smaller images determined by the location information that is supplied by the bounding boxes. After that, a method called boundary identification is used to determine the border parameters of each loess landslide that has been found. This helps to enhance the effectiveness of the segmentation procedure. Because additional inception blocks with dilatation were included in the construction of the segmentation U-Net, its effectiveness in landslide segmentation has significantly increased as a direct consequence. It is well known that separating loess landslides into their parts is a difficult task. This is mostly attributable to the intrinsic qualities of boundary information that is ambiguous. The novel framework is tested in comparison to the conventional U-Net as well as other modern benchmark landslide segmentation methods. The results of the computer analysis show that the suggested structure achieves a level of accuracy in dividing up loess landslides that is much higher than that achieved by the other benchmarking methods that were looked at.