A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Combining BotNet and ResNet Feature Maps for Accurate Landslide Identification Using DeepLabV3+
The field of landslide research involves two aspects: landslide identification and risk assessment of landslide disasters, with landslide identification forming the basis for all landslide studies. With the development of computer vision, the application of deep learning to landslide identification research is also gradually expanding. This paper proposes an improved deeplabV3+ landslide identification method that combines BotNet and ResNet feature maps through a feature fusion module, which pays attention to both local and global features. Comparative experiments and ablation experiments were conducted on a dataset containing 1227 landslide images. The dataset was divided into a training set and a validation set in a ratio of 9:1. The experimental results show that the proposed optimization method performs the best on the validation set, with an average intersection over union (mIoU) of 82.50%, pixel accuracy (PA) of 93.63%, recall rate of 84.59%, and mean class pixel accuracy (mCPA) of 89.95%.
Combining BotNet and ResNet Feature Maps for Accurate Landslide Identification Using DeepLabV3+
The field of landslide research involves two aspects: landslide identification and risk assessment of landslide disasters, with landslide identification forming the basis for all landslide studies. With the development of computer vision, the application of deep learning to landslide identification research is also gradually expanding. This paper proposes an improved deeplabV3+ landslide identification method that combines BotNet and ResNet feature maps through a feature fusion module, which pays attention to both local and global features. Comparative experiments and ablation experiments were conducted on a dataset containing 1227 landslide images. The dataset was divided into a training set and a validation set in a ratio of 9:1. The experimental results show that the proposed optimization method performs the best on the validation set, with an average intersection over union (mIoU) of 82.50%, pixel accuracy (PA) of 93.63%, recall rate of 84.59%, and mean class pixel accuracy (mCPA) of 89.95%.
Combining BotNet and ResNet Feature Maps for Accurate Landslide Identification Using DeepLabV3+
Wan, Yixuan (author) / Huang, Jianhua (author) / Ji, Yuanfa (author) / Yu, Zhengyao (author) / Luo, Mingming (author)
2023-05-26
862050 byte
Conference paper
Electronic Resource
English
Semantic Segmentation of Cracks Using DeepLabv3+
| Springer Verlag | 2023
Identification of Landslide Vulnerable Slopes using Topographical Maps
| British Library Conference Proceedings | 2002
Segmentation of Concrete Surface Cracks Using DeeplabV3+ Architecture
| Springer Verlag | 2023