Evidential Fusion Based Technique for Detecting Landslide Barrier Lakes From Cloud-Covered Remote Sensing Images

Abstract

Landslide barrier lakes usually form quickly after disasters and require very timely remote sensing images to monitor the land-cover change. However, cloud-free images are not always available in emergency situations. This paper provides a method to fuse multitemporal cloud-covered images for change detection, based on the evidential fusion framework. First, the frame of discernment is defined by postclassification comparison results. Second, a way of measuring the basic belief assignment (BBA) is introduced based on the confusion matrixes. Next, a simple BBA redistribution process is proposed to deal with cloud coverage problems. Then, the complementary and redundant information from the input images can be fused following the evidence combination and decision making rules in the evidential fusion framework. Finally, the land-cover change map can be derived. Thanks to the Dempster–Shafer evidence theory, the proposed method can complete the change detection process—including data fusion and cloud removal—in an integrated manner. The proposed method is applied to detect the landslide barrier lake in a real case study, using a series of cloud-covered images from the GF-1 satellite. Result comparisons show that the proposed method is more effective than some basic fusion strategies that perform change detection and cloud removal in separate steps. Then, some approaches to improve the proposed method are discussed: introducing new evidence combination rule, improving the classification accuracy, and adding new evidences. All the results indicate the potential of evidential fusion for change detection from cloud-covered images.