基于深度学习的高分辨率遥感影像滑坡体识别方法研究

1.成都大数据产业技术研究院有限公司,成都 610095;2.四川天府新区创新装备研究院,成都 610299;3.重庆大学自动化学院,重庆 400044;4.北京理工大学,北京 100081

深度学习;滑坡检测;卷积神经网络;特征提取;损失函数

Research on Landslide Detection in High-Resolution Remote Sensing Image Based on Deep Learning
CHEN Long1,GE Cheng1,DAI Yingchao2,3,WANG Hongyu4,LIU Weiwei2

1.Chengdu Big Data Industry Technology Research Institute Co.,Ltd.,Chengdu 610095,China;2.Innovative Equipment Research Institute of Beijing Institute of Technology in Sichuan Tianfu New Area , Chengdu 610299, China;3.School of Automation , Chongqing University , Chongqing 400044, China;4.Beijing Institute of Tech⁃nology,Beijing 100081,China

Deep learning;Landslide detection;Convolutional neural networks;Feature extraction;Loss func⁃tion

DOI: 10.13512/j.hndz.2025.02.09

备注

针对现有滑坡体检测精度低的现实问题,提出了一种基于深度学习的滑坡体检测框架。该框架包含数据采集与处理、特征选择、检测模型三个部分,可融合多源数据有效提高对滑坡体检测能力。提出了多模态的KlA⁃lexNet模型,可实现像素级分割预测,有效地融合空间特征。实验结果表明,所提KlAlexNet模型具备高精度的滑坡体检测能力。与U-Net、U-Net++、FC_DenseNet、YOLOv9-seg等方法相比具备优势。实验结果验证了所提方法的有效性和实用性,该方法具有广阔的应用前景。
In response to the current issue of low precision in landslide detection,a deep learning-based landslide detection framework was proposed. This framework includeed three parts:data collection and processing,feature selection,and detection model,which can effectively improve the detection capability of landslides by integrating multi-source data. A multimodal KlAlexNet model was proposed,which could achieve pixel-level segmentation pre⁃diction and effectively fuse spatial features. Experimental results indicate that the proposed KlAlexNet model has high precision in landslide detection. Compared with methods such as U-Net,U-Net++,FC_DenseNet,and YO⁃LOv9-seg, it demonstrates advantages. The experimental results validate the effectiveness and practicality of the proposed method,indicating its broad application prospects.
·