Recognition of landslide triggers in southeast Tibetan (China) using a novel lightweight network.

The Tibetan Plateau is a driver and amplifier of global climate change. The increased frequency and scale of landslides in this area are one of the manifestations of extreme climate change. Studying the trigger of landslides is of great value to the research, protection, and management of engineering geology and climatic environmental changes. However, to our knowledge, there is no efficient, convenient and intelligent method to recognize the trigger of landslides in the Tibetan Plateau. Therefore, a new high-efficiency and high-precision deep learning algorithm has been proposed in this study to analyze the landslides triggers. Specifically, this paper proposes a novel lightweight neural network landslide classification method (MNTL) based on MobileNet-V2 and transfer learning. Mobilenet-V2 requires few parameters and few floating-point operations per second. Furthermore, it is integrated with transfer learning to improve the representation learning ability of the model. The proposed method was applied to classify landslides induced by rainfall and thawing effect. The method required only 30 samples of each class, and it converged quickly in just 3 min and achieved a 94% forecast accuracy. Compared with six state-of-the-art deep learning classification methods—specifically, VGG-16, VGG-19, ResNet-50, ResNet-101, Inception, and MobileNet-V2, the proposed method exhibited competitive advantages in terms of convergence speed and generalization capability. MNTL can be embedded into the mobile terminal, which is conducive to the rapid application of landslide-related research results at a small cost. More importantly, the work in this paper could serve as a potential basis for advancing research on the correlation between landslide hazards and climate change in southeast Tibet. [ABSTRACT FROM AUTHOR]