Jin Yuanhang, Liu Xiaosheng, Huang Xiaobin
School of Civil Engineering and Surveying & Mapping Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, China.
Sensors (Basel). 2024 Jun 6;24(11):3677. doi: 10.3390/s24113677.
Landslides constitute a significant hazard to human life, safety and natural resources. Traditional landslide investigation methods demand considerable human effort and expertise. To address this issue, this study introduces an innovative landslide segmentation framework, EMR-HRNet, aimed at enhancing accuracy. Initially, a novel data augmentation technique, CenterRep, is proposed, not only augmenting the training dataset but also enabling the model to more effectively capture the intricate features of landslides. Furthermore, this paper integrates a RefConv and Multi-Dconv Head Transposed Attention (RMA) feature pyramid structure into the HRNet model, augmenting the model's capacity for semantic recognition and expression at various levels. Last, the incorporation of the Dilated Efficient Multi-Scale Attention (DEMA) block substantially widens the model's receptive field, bolstering its capability to discern local features. Rigorous evaluations on the Bijie dataset and the Sichuan and surrounding area dataset demonstrate that EMR-HRNet outperforms other advanced semantic segmentation models, achieving mIoU scores of 81.70% and 71.68%, respectively. Additionally, ablation studies conducted across the comprehensive dataset further corroborate the enhancements' efficacy. The results indicate that EMR-HRNet excels in processing satellite and UAV remote sensing imagery, showcasing its significant potential in multi-source optical remote sensing for landslide segmentation.
山体滑坡对人类生命、安全和自然资源构成重大危害。传统的山体滑坡调查方法需要大量人力和专业知识。为了解决这个问题，本研究引入了一种创新的山体滑坡分割框架EMR-HRNet，旨在提高准确性。首先，提出了一种新颖的数据增强技术CenterRep，它不仅扩充了训练数据集，还使模型能够更有效地捕捉山体滑坡的复杂特征。此外，本文将RefConv和多卷积头转置注意力（RMA）特征金字塔结构集成到HRNet模型中，增强了模型在各个层次上的语义识别和表达能力。最后，引入扩张高效多尺度注意力（DEMA）模块大大拓宽了模型的感受野，增强了其辨别局部特征的能力。在毕节数据集以及四川及周边地区数据集上进行的严格评估表明，EMR-HRNet优于其他先进的语义分割模型，分别实现了81.70%和71.68%的平均交并比（mIoU）分数。此外，在综合数据集上进行的消融研究进一步证实了这些改进的有效性。结果表明，EMR-HRNet在处理卫星和无人机遥感图像方面表现出色，在多源光学遥感山体滑坡分割中具有巨大潜力。
