赵怡晴, 黄晢航, 刘宏发, 金爱兵, 陆通, 刘金博. 露天矿边坡裂隙智能识别与信息解算[J]. 工程科学学报. DOI: 10.13374/j.issn2095-9389.2023.07.31.002
引用本文: 赵怡晴, 黄晢航, 刘宏发, 金爱兵, 陆通, 刘金博. 露天矿边坡裂隙智能识别与信息解算[J]. 工程科学学报. DOI: 10.13374/j.issn2095-9389.2023.07.31.002
Intelligent identification and information calculation of slope crack in open-pit mine[J]. Chinese Journal of Engineering. DOI: 10.13374/j.issn2095-9389.2023.07.31.002
Citation: Intelligent identification and information calculation of slope crack in open-pit mine[J]. Chinese Journal of Engineering. DOI: 10.13374/j.issn2095-9389.2023.07.31.002

露天矿边坡裂隙智能识别与信息解算

Intelligent identification and information calculation of slope crack in open-pit mine

  • 摘要: 节理裂隙是影响露天矿边坡稳定性的重要因素之一,为快速获取节理裂隙几何信息,提出一种露天矿边坡裂隙识别及几何参数解译方法。利用由无人机航拍露天矿边坡裂隙图像,采用随机旋转、随机亮度及对比度调整等方式进行数据增广形成裂隙图像数据集;采用残差网络(ResNet)18、34、50、101和152五种模型对U-net网络的骨干架构网络进行改进,提出基于改进U-net网络的边坡裂隙识别模型;基于像素二分类问题采用准确率(Accuracy)、交并比(IoU)和F1分数(F1 Score)作为评价指标,采用裂隙图像数据集对提出模型进行训练和评估,输出裂隙二值图,并与传统裂隙识别方法识别结果进行对比;对裂隙二值图进行裂隙几何参数信息解算,获得裂隙长度、宽度统计分布规律和参数。结果表明:ResNet模型对U-net模型改进可以提高模型的评价指标,且随着边坡裂隙识别模型的骨干网络层次加深性能提高,Res101-Unet模型的Accuracy、IoU、F1 Score分别达到了95.12%、60.13%、79.53%,同时也高于传统识别算法的相应指标。

     

    Abstract: Nodal cracks are one of the important factors affecting the stability of slopes in open pit mines. In order to quickly obtain geometric information of nodal cracks, a method of identifying and deciphering geometric parameters of slope cracks in open pit mines is proposed. Using aerial photography of open pit slope crack images by UAV, the data is expanded by random rotation, random brightness and contrast adjustment to form a crack image data set; five models of residual network (ResNet) 18, 34, 50, 101 and 152 are used to improve the backbone architecture network of U-net network, and a slope crack recognition model based on the improved U-net network is proposed The proposed model is trained and evaluated using the crack image dataset, and the crack binary map is output and compared with the recognition results of traditional crack recognition methods; the crack binary map is solved for the crack geometric parameter information to obtain the crack length, width and parameters, The crack geometric parameters are solved to obtain the statistical distribution patterns and parameters of crack length and width. The results show that the improvement of the ResNet model to the U-net model can improve the evaluation index of the model, and the performance of the Res101-Unet model improves with the deepening of the backbone network level of the slope crack identification model, and the Accuracy, IoU, and F1 Score of the Res101-Unet model reach 95.12%, 60.13%, and 79.53%, respectively, and also higher than the corresponding index of the traditional identification The corresponding indexes of the algorithms are also higher than those of the traditional recognition algorithms.

     

/

返回文章
返回