基于图像深度特征学习的矿山井筒设施隐患检测方法研究

Research on Hidden Danger Detection Method of Mine Shaft Facilities Based on Deep Image Feature Learning

针对人工完成矿山井筒设施检测时面临成本高和安全威胁等问题,提出利用采集的井筒设施监控视频图像,采用基于图像深度特征学习的矿山井筒设施隐患检测方法对井筒设施进行隐患检测。构建包含大量正常设施图像和少量隐患图像的数据集;同时构建包含特征提取网络和分类网络的并行神经网络作为训练网络,在正常图像集、外部图像集和少量隐患图像数据集上,设计联合损失函数对网络进行训练;通过聚类策略选择具有代表性的正常图像深度特征作为特征模板,通过模板匹配实现隐患图像检测。试验结果表明,在构建的矿山井筒井架设施图像数据集上,所提算法训练的特征提取模型能有效提高图像特征的表达能力,模型的隐患检测精度可达到92%以上,为利用矿山井筒设施图像进行矿山井筒的隐患检测提供了新思路。

Aiming at the problems of high'cost and safety threat when manually detecting the facilities in the mine shaft,a hidden danger detection method of mine shaft facilities based on deep image feature learning(HDD-DIFL)was proposed by using the sampled video images of the mine shaft facilities.Firstly,a dataset containing a large number of normal facility images and a small number of hidden danger images was built.Then,a parallel neural network including feature extraction network and classification network was constructed as a training network.A joint loss function was designed to train the network on the normal image set,the external image set and a small number of hidden danger image dataset.Finally,some representative normal deep features were selected as the feature templates by clustering strategy,and the hidden danger facility images were detected by template matching.The experimental results show that the feature extraction model trained by the proposed algorithm can effectively improve the expression ability of image features on the constructed image dataset of mine shaft facilities,and the hidden danger detection accuracy of the model can reach more than 92%.The study provides a new thinking for the hidden danger detection of the mine shaft by using the sampled facility videos.