摘要
近年来,滑坡灾害频繁发生,灾后及时对滑坡进行识别对减灾和灾后重建具有重要意义。在高分辨率遥感影像滑坡识别任务中,普遍存在有正负样本不均衡、特征提取不完善的问题。为此,本文提出一种基于HRNet的高分辨率遥感影像滑坡识别模型,该模型的高分辨率特征金字塔网络结构,在多尺度下充分捕捉地质特征,提高模型对复杂地形的适应性。同时,在训练过程中采用FocalLoss损失函数,为难以分类的正样本赋予更大的权重,有效缓解了正负样本不均衡问题,这使得模型更加关注难以区分的滑坡区域,进一步提升了滑坡识别的准确性和鲁棒性。实验结果表明,该方法在数据集中实现了优越的滑坡分割性能,对灾后的快速反应有指导性意义。
In recent years,landslides have occurred frequently.It is of great significance to identify the landslide in time for disaster reduction and post-disaster reconstruction.In the landslide recognition task of high resolution remote sensing image,there are problems of uneven positive and negative samples and imperfect feature extraction.In this paper,a high-resolution remote sensing image landslide recognition model based on HRNet is proposed.The high-resolution feature pyramid network structure of the model can fully capture geological features at multiple scales and improve the model's adaptability to complex terrain.At the same time,Focal Loss loss function was adopted in the training process to assign greater weight to positive samples that were difficult to be classified,which effectively alleviated the problem of imbalance between positive and negative samples.The model pays more attention to the difficult to distinguish landslide area,which further improves the accuracy and robustness of landslide identification.The experimental results show that the proposed method achieves superior landslide segmentation performance in the data set,and has guiding significance for the rapid response after disaster.
作者
何杰
HE Jie(Sichuan Real Estate Registration Center,Chengdu 610072;Southwest Jiaotong University,Chengdu 611756)
出处
《四川地质学报》
2024年第3期567-570,共4页
Acta Geologica Sichuan
关键词
滑坡
高分辨率遥感图像
语义分割
landslide
high-resolution remote sensing images
semantic segmentation
