Intelligent Recognition Of Landslide Boundaries And Rock Crevices Based On Convolutional Neural Networks

By the development of AI,the combination of deep learning and geotechnical engineering has gradually become a research hotspot.In this paper,we study convolutional neural networks in deep learning and their optimization algorithms,and combine them with the study of landslide boundary detection and parameter extraction of rock fractures in geotechnical engineering.The specific work mainly includes.(1)A brief review of image segmentation techniques in computer vision,etc.,and an introduction of deep learning related theories,frameworks and improved algorithms,which provide theoretical and data support for subsequent landslide boundary detection and parameter extraction of rock fissures.(2)The optimization algorithm of landslide boundary recognition based on double attention mechanism is established.To address the problems of simple network architecture,low landslide recognition accuracy and low segmentation accuracy caused by a single landslide dataset in current domestic and international research,a landslide image dataset with multiple data sources is established,and the landslide boundary detection model is trained and tested with this dataset to increase the model prediction accuracy and model robustness.Ten network architectures are compared and analyzed,and the Res-101 network architecture is selected as the backbone network of the landslide boundary detection model.At the same time,the information overload problem is solved by introducing the attention mechanism to focus on the key information of the image,which increases the efficiency of landslide image processing and improves the accuracy of landslide boundary segmentation.The experiments prove that the algorithm of this paper has certain advantages in terms of arithmetic power and speed,and has better recognition effect for different types of landslides,and the landslide boundary segmentation results are closer to the real landslide boundary.(3)To address the problem that there are few studies combining deep learning for complex fracture network identification and single fracture extraction and characterization,a high-precision open-air slope aerial photograph rock fracture dataset was established based on UAV aerial survey system to increase the robustness of the segmentation model.A new fracture clipping and splicing method is proposed based on the improved U-net structure using cavity convolution and DUpsampling methods to obtain fracture images with higher segmentation accuracy,which improves the accuracy of fracture segmentation and parameter extraction compared with traditional methods.