摘要
针对地下煤炭资源被开采利用,出现大规模采空区,产生大面积沉陷,威胁到矿区的安全生产和人们的人身安全问题。本研究首先对鹤岗矿区遥感影像进行预处理、数据集的制作,然后用U-Net卷积神经网络训练模型,修改训练参数,进行塌陷坑、地裂缝、塌陷盆地的识别,整体精度达到97.58%,表明此次模型训练结果较准确。从而得到鹤岗矿区地裂缝、塌陷坑、塌陷盆地的具体分布图,发现盆地的查准率最高,地裂缝识别效果其次,塌陷坑由于本身形状及特征的因素影响,识别的效果差于其他两类地物,分析上述三类地物的精度指标,证明了U-Net网络模型适合于本次研究的研究,为复垦、治理采空塌陷区提供基础资料具有重要的意义。
Due to the mining and utilization of underground coal resources,large-scale goaf areas have emerged,causing extensive subsidence and posing a threat to the safety of mining operations and personnel.This article initially preprocesses remote sensing images of the Hegang mining area and compiles a dataset.Subsequently,the U-Net convolutional neural network is employed to train the model,adjust training parameters,and identify subsidence pits,ground fissures,and subsidence basins.The overall accuracy rate of 97.58%demonstrates the relatively high precision of the model training results.Consequently,detailed distribution maps of ground fissures,subsidence pits,and subsidence basins in the Hegang mining area have been obtained.It was discovered that the basin had the highest precision,followed by the recognition of ground fissures.Due to their inherent shape and characteristics,the recognition of subsidence pits was less accurate compared to the other two land features.Analyzing the accuracy indicators of these three land features confirms that the U-Net network model is suitable for this study and provides valuable basic data for the reclamation and management of goaf subsidence areas,which is of significant importance.
作者
杨书平
Yang Shuping(Zhengzhou University of Science and Technology,Zhengzhou,China)
出处
《科学技术创新》
2025年第18期27-30,共4页
Scientific and Technological Innovation
关键词
U-Net
深度学习
塌陷区识别
采煤塌陷区
U-Net
deep learning
mining area monitoring
subsidence area recognition
