摘要
随着遥感技术的逐步成熟,滑坡目标分割技术在灾害防控、城市建设等领域发挥着越来越关键的作用。然而,面临着卷积神经网络在理解上下文语义信息上存在困难以及滑坡数据集中正负样本分布不均的问题,需要寻找一种更优的解决方案。为此,提出了一种改进Swin Transformer的滑坡分割算法。首先,将Swin Transformer引入到滑坡目标分割任务中,并构建了Swin-Unet网络。其次,为了提高网络的表征能力,在Swin-Unet中加入通道注意力模块;为了平衡小样本正负样本不均衡,使用在线困难样例挖掘;为了进一步提高模型性能,引入并优化了条件随机场。实验结果表明,相较于基线算法,改进方法在滑坡目标分割任务上取得了更优的性能,Dice提高了1.9%,交并比(IoU)提高了1.2%,研究为滑坡目标分割提供了一种有效的方法,并为相关领域的研究提供了新的思路。
As remote sensing technology continues to mature,landslide segmentation techniques are playing an increasingly critical role in disaster prevention and urban development.However,convolutional neural networks face challenges in understanding contextual semantic information,and there is an imbalance in the distribution of positive and negative samples in landslide datasets,necessitating a superior solution.To address these issues,this study proposes a landslide segmentation algorithm based on the improved Swin Transformer.Initially,the study incorporates the Swin Transformer into the landslide segmentation task and constructs a Swin-Unet network.Subsequently,to enhance the network's representational capability,a channel attention module is added to the Swin-Unet.To balance the imbalanced positive and negative minor samples,online hard example mining is utilized.To further improve the performance of the model,a conditional random field was introduced and optimized.Experimental results show that,compared to the baseline algorithm,the improved method proposed in this study achieves superior performance on the landslide segmentation task,with a 1.9%increase in the Dice index and a 1.2%increase in the IoU index.This research provides an effective method for landslide segmentation and offers a new perspective for research in related fields.
作者
张思远
谭晓玲
朱木雷
杨伟良
Zhang Siyuan;Tan Xiaoling;Zhu Mulei;Yang Weiliang(School of Electronic and Information Engineering,Chongqing Three Gorges University,Chongqing 404100,China;Key Laboratory of Geological Environment Monitoring and Disaster Early Warning in the Three Gorges Reservoir Area,Chongqing 404100,China)
出处
《国外电子测量技术》
北大核心
2023年第11期49-56,共8页
Foreign Electronic Measurement Technology
基金
三峡库区地质环境监测与灾害预警重庆市重点实验室开放基金(ZD2020A0302)项目资助。
