Research on automatic recognition of active landslides using InSAR deformation under digital morphology: A case study of the Baihetan reservoir, China

Optical remote sensing and field investigations cannot satisfy the accuracy and timeliness requirements of active landslide detection. Interferometric synthetic aperture radar (InSAR) technology has become the mainstream method for observing active landslides in recent years, due to its advantages of a large detection range and high sensitivity to surface deformation. However, quickly and accurately obtaining landslide boundaries from InSAR results is still a key issue for hazard mitigation and watershed management. In this study, first, an automatic recognition method for active landslides based on InSAR results is established to rapidly extract deformed slopes. In the recognition process, using the deformation value map as the object, the optimal threshold is determined using the image gradient edge information to extract the deformed pixels. Second, the contrast limited adaptive histogram equalization (CLAHE) algorithm is employed to improve the contrast of images with weak deformation. Last, morphological rules are applied to optimize the segmentation results to ensure that they are near the boundary of the natural landslide. To verify the effectiveness of the method, the Baihetan Reservoir area, which has frequent landslides in the lower reaches of the Jinsha River Basin, was selected as the test area. Under the conditions of ascending and descending orbits, 336 and 590 landslides, respectively, were recognized. Through unmanned aerial vehicle (UAV) and field investigations, the recognition accuracy of 76% is reached without sample training, and the maximum intersection over union (IOU) of a single landslide is increased by 0.3. This finding shows that the automatic recognition method can quickly identify dangerous active landslides at large spatial scales and with complex topographies.