Due to the seasonal permafrost thawing, the Qinghai–Tibet Highway has a depression and instability of the roadbed. In order to obtain the ablation interface and water content characteristics of seasonal permafrost areas, non-destructive ground penetrating radars using electromagnetic wave detection methods are widely used. Regarding the imaging of the ablation interface in permafrost regions, this paper proposes a high-precision procedure for seasonal permafrost media using waveform difference analysis, electromagnetic wave attenuation attribute calculation and relative wave impedance conversion. It improves the resolution and division accuracy of the imaging. In addition, the study demonstrates a method to calculate the temperature and water content of the ablation zone by mining attenuation attribute, relative wave impedance attribute, absolute instantaneous amplitude attribute and the weighted average frequency attribute parameters under the constraints of the measured data. This method has high accuracy and high efficiency and can be used in the rapid calculation of temperature and water content of seasonal permafrost on the Qinghai–Tibet Highway.

The ‘intention’ classification of a user question is an important element of a task-engine driven chatbot. The essence of a user question’s intention understanding is the text classification. The transfer learning, such as BERT (Bidirectional Encoder Representations from Transformers) and ERNIE (Enhanced Representation through Knowledge Integration), has put the text classification task into a new level, but the BERT and ERNIE model are difficult to support high QPS (queries per second) intelligent dialogue systems due to computational performance issues. In reality, the simple classification model usually shows a high computational performance, but they are limited by low accuracy. In this paper, we use knowledge of the ERNIE model to distill the FastText model; the ERNIE model works as a teacher model to predict the massive online unlabeled data for data enhancement, and then guides the training of the student model of FastText with better computational efficiency. The FastText model is distilled by the ERNIE model in chatbot intention classification. This not only guarantees the superiority of its original computational performance, but also the intention classification accuracy has been significantly improved.

Ground-penetrating radar (GPR) is often used to detect targets in a construction environment. Due to the different construction environments, the noise exhibits different characteristics on the GPR signal. When the noise is widely distributed on the GPR signal, and its spectrum and the spectrum of the active signal are aliased, it is difficult to separate and suppress the noise by traditional filtering methods. In this paper, we propose a deep learning GPR image noise suppression method based on a recursive guided and dual multi-scale self-attention mechanism neural network (RG-DMSA-NN), which uses a recursive guidance module and a dual multi-scale self-attention mechanism module to improve the feature extraction ability of the image and enhance the robustness and generalization ability in image noise suppression. Through the application of noise suppression on the synthesized test data and the GPR data actually collected by the Macao Science and Technology Museum, the advantages of this method over the traditional filtering, DnCNN and UNet noise suppression methods are demonstrated.

Due to the impact of climate warming and engineering construction, thaw-slumping has developed extensively along the Qinghai–Tibet Project Corridor. These landslide disasters not only destroy the fragile ecology of the Qinghai–Tibet Plateau but also threaten the security of the Qinghai–Tibet Project Corridor. Because remote-sensing images lack imaging data inside the landslide body, and the excavation of boreholes has blindness and inefficiency, the ground-penetrating radar method with high efficiency and deep imaging has been developed and applied in the detection and treatment of thaw-slumping. To more accurately divide the soil-layered structure of the thaw-slumping body and obtain the key elements of the thaw-slumping such as temperature change trend and relative water content, we propose the use of amplitude event axis tracking and amplitude energy attenuation calculation to divide the fine layering of the thaw-slumping body. In addition, based on layer division, we introduce two attribute parameters to participate in the calculation of relative water content. These two attribute parameters are the weighted average frequency attribute, which reflects the temperature change trend, and the sweetness attribute, which reflects the change in the physical properties of the underground medium. The calculated 3D profile and time slice of the relative water content comprehensively show the change characteristics and enrichment area of the internal relative water content of the thaw-slumping. These methods and results are valuable for the characterization, evaluation, and treatment of thaw-slumping.