深度学习在矿产资源卫片执法中的应用研究

The Application of Deep Learning in Law Enforcement of Mineral Resources with Remote Sensing Technology

矿产资源卫片执法是资源监管与保护的重要技术手段,而深度学习是当前人工智能领域中的研究热点,为提高卫片执法效率、降低人力解译工作量提供了可能。本文提出了一种基于Mask RCNN的遥感影像露天矿山疑似违法图斑自动检测提取方法,通过采集、扩充、规范矿山样本,制作特定的coco数据集输入到Mask RCNN进行有监督学习训练。利用训练出的分类网络模型进行遥感影像矿山图斑的自动提取,并以采矿权矢量图层作为判定依据,自动圈取影像中的疑似违法图斑。经实验,本文方法的mAP精度达87%以上,高于传统方法,对深度学习在卫片执法中的应用研究作出了有效实践。

Law enforcement with remote sensing technology is an important technical means for mineral resources supervision and protection. Deep-learning is a hotspot in the field of artificial intelligence and provides a possibility for improving the efficiency of law enforcement and the manual extraction of information. Based on Mask RCNN algorithm, this paper proposes an automatic method to extract the potentially illegal surface mine in remote sensing images. Firstly, the specific coco dataset which is made by manually extracted extended formal specimens is input to Mask RCNN model for supervised learning training. Secondly, the surface mine information is extracted by this trained model. Judging by the Vector layer of mining right permit, the potentially illegal surface mine is identified automatically. After experiment, the mAP of this method is more than 87%, and it has a better effect than KNN, SVM. Finally, we are able to propose this method which leads to effective research in the application of deep-learning to law enforcement with remote sensing technology.