矿山遥感监测实例分割数据集的自动生成方法

Automatic generation method of instance segmentation data set in mine remote sensing monitoring

为解决基于遥感影像的实例分割数据集匮乏和人工标注效率低、工作量大的问题,在已有矿山遥感监测数据的基础上,通过基于ArcGIS二次开发组件库和JSON(JavaScript object notation)操作类库研究了矿山遥感监测遥感影像属性数据转换方法及其矢量数据与JSON文件的转换问题,并将自动生成的矿山遥感监测实例分割数据集应用于Mask-RCNN网络模型。实验证明:通过矿山遥感监测矢量数据自动生成实例分割数据集与实际解译数据的精度一致;该方法生成的数据集可作为深度学习网络模型的训练集,提高标注效率;在保证数据准确性的情况下,可实现对解译数据的再利用,并高效生成规范的深度学习数据集。

In order to solve the problem of few instance segmentation data set based on remote sensing images,low efficiency of manual annotation and heavy workload of manual control,a conversion method between the remote sensing vector data and JSON(JavaScript object notation)files were studied through the remote sensing image processing based on ArcGIS secondary development component library.It was based on the existing data of mine remote sensing monitoring and JSON operation class library.The automatically generated mine remote sensing monitoring’s instance segmentation data set was applied to the Mask-RCNN network model.The results show that there is no deviation between the accuracy of the mine remote sensing monitoring’s instance segmentation data and the interpretation data.The data set generated by this method can be used as the training set of deep learning network model,and the efficiency of the proposed method is obviously higher than that of manual annotation.If the accuracy of the data is ensured,the method can help to realize the reue of the interpretation data and efficiently generate the standard deep learning data set.