Landslide Mapping Using Two Main Deep-Learning Convolution Neural Network Streams Combined by the Dempster–Shafer Model

Beyond the direct hazards of earthquakes, deposited mass earthquake-induced landslide (EQIL) in riverbeds causes river to thrust upward. The EQIL inventories are generated mostly by traditional or semisupervised mapping approaches, which required a parameter's tuning binary threshold decision practical application. In this study, we investigated impact optical data from PlanetScope sensor and topographic factors ALOS on using deep-learning convolution neural network (CNN). Thus, six training datasets were prepared used evaluate performance CNN model only these along with each all across west coast Trishuli Nepal. For first time, Dempster-Shafer (D-S) was applied for combining resulting maps stream that trained different datasets. Finally, seven compared against detailed accurate inventory polygons mean intersection-over-union (mIOU). Our results confirm dataset spectral information factor slope is helpful distinguish bodies other similar features, such as barren lands, consequently increases accuracy. improvement mIOU range approximately zero more than 17%. Moreover, D-S can be considered an optimizer method combine scenarios.