基于Gabor域时空泛化建模的矿井水灾感知方法

Mine flood perception method based on spatiotemporal generalization modeling in Gabor domai

  • 摘要: 矿井突涌水图像中突涌水纹理与煤岩图像相比具有时空域变化性强的特点,现有基于图像纹理特征的矿井水灾识别方法对于复杂突涌水纹理特征的提取能力有限、识别率较低。针对该问题,提出了一种基于Gabor域时空泛化建模的矿井水灾感知方法。该方法分别对训练样本图像和待测样本图像进行不同感受野、不同方向下的Gabor分解,将各子带的期望与标准差组合,构成本方向的学习特征向量和待测特征向量;根据最小熵原理对特征向量进行时空泛化建模,以去除时空域敏感性;采用特征向量各分量之间的夹角作为相似性测度,对学习特征向量和待测特征向量进行相似性比较,实现突涌水识别。实验结果表明,该方法识别率达89.4%,识别时间为136 ms,基本满足井下水灾实时感知需求。

     

    Abstract: Compared with coal and rock images, water inrush texture in mine water inrush images has strong variability characteristics in spatiotemporal domain. Existing mine flood identification methods based on image texture features have limited extraction ability and low recognition rate for complex water inrush texture features. For the above problems, a mine flood perception method based on spatiotemporal generalization modeling in Gabor domain is proposed. In the method, Gabor decompositions of training sample images and tested sample images under different receptive fields and directions are carried out separately, and expectation and standard deviation of each sub-band are combined to form learning feature vector and the tested one in a direction. Spatiotemporal generalization modeling of the feature vectors is carried out according to the minimum entropy principle, so as to remove spatiotemporal sensitivity. Angles between each component of the feature vector are taken as the similarity measure of similarity comparison between the learning feature vector and the tested one, so as to realize water inrush recognition. The experimental results show that recognition rate of the method is 89.4% and recognition time is 136 ms, which basically meets real-time perception demand of mine flood.

     

/

返回文章
返回