摘要
在黄土滑坡识别的研究中,影像数据集通常规模较小,并且滑坡区域的光谱特征与周围环境相似,致使现有滑坡自动识别算法易出现误判和漏判的情况。使用甘肃省黑方台滑坡影像制作黄土滑坡影像数据集,并在YOLOv5模型的基础上,将Alpha-EIoU损失函数代替为GIoU损失函数,引入ECA注意力通道机制和解耦合检测头,提出一种适用于黄土滑坡识别YOLOv5-LD改进模型。实验结果表明,YOLOv5-LD模型识别黑方台滑坡的Precision为85.0%,Recall为64.9%,AP为77.1%,F1-Score为73.6%,与原模型YOLOv5相比,Recall、AP、F1-Score分别提升9.5%、5.0%、5.8%,并且优于目前主流的几种深度学习模型的识别性能。在可视化实验中,YOLOv5-LD漏检率低,识别效果较好,对于小数据集黄土滑坡具有较好的识别能力,可为深度学习方法在黄土滑坡识别中的应用研究提供参考。
Accurate and effective landslide identification is the foundation of landslide disaster prevention and control.In the studies of loess landslide identification,the image datasets are usually small in size and the spectral features of the landslide area are similar to the surrounding environment,which makes the existing automatic landslide identification algorithms prone to misclassification and underclassification.This article uses the image of Heifangtai landslide in Gansu Province to produce a loess landslide image dataset,and on the basis of YOLOv5 model,this article replaces the Alpha-EIoU loss function with GIoU loss function,introduces the ECA attention channel mechanism and decoupled head,and proposes an improved YOLOv5 model suitable for the identification of loess landslides—YOLOv5-LD.The experimental results show that the precision(P)of YOLOv5-LD model for recognizing the landslide of Heifangtai is 85.0%,the recall(R)is 64.9%,the average precision(AP)is 77.1%,and the F1-Score is 73.6%,which compare with the original model YOLOv5,the R,AP,and F1-Score are improved by 9.5%,5.0%,and 5.8%,respectively,and outperform the recognition performance of several current mainstream deep learning models.In the visualization experiments,YOLOv5-LD has a low leakage rate and a better recognition effect,and has a better recognition ability for loess landslides with small datasets,which provides a valuable reference for future research on deep learning methods in loess landslide recognition.
作者
王浩男
王利
舒宝
赵丽华
瞿伟
许豪
WANG Haonan;WANG Li;SHU Bao;ZHAO Lihua;QU Wei;XU Hao(College of Geology Engineering and Geomatics,Chang’an University,Xi’an 710054 China;Key Laboratory of Loess,Xi’an 710054,China;Key Laboratory of Western China’s Mineral Resource and Geological Engineering,Ministry of Education,Xi’an 710054,China;Key Laboratory of Ecological Geology and Disaster Prevention,Ministry of Natural Resources,Xi’an 710054,China)
出处
《测绘工程》
2025年第6期64-71,80,共9页
Engineering of Surveying and Mapping
基金
国家重点研发计划资助项目(2024YFC3012603)
陕西省科技创新团队资助项目(2021TD-51)
陕西省地学大数据与地质灾害防治创新团队资助项目(2022)。
关键词
黄土滑坡
滑坡识别
深度学习
损失函数
loess landslide
landslide recognition
deep learning
loss function
