基于图像纹理特征的矿井水灾感知方法

现有水灾监测报警方法存在着适应性差、误报和漏报率高等问题,难以满足煤矿安全生产需求。为此,提出了基于图像纹理特征的矿井水灾感知方法:在巷道顶部或巷帮、采煤工作面支架等设置摄像机,实时采集采掘工作面和巷道底板图像;通过双树复小波变换,提取水、煤、岩石图像纹理特征,建立水灾图像识别模型;对学习样本进行双树复小波变换,提取1,2级系数,统计其方差与期望值,利用相应的方差与期望值构造泊松分布模型,并估计其各个方向模型的强度参数。对待测样本同样进行双树复小波变换后,利用1,2级系数的方差与期望值构成的模型强度参数向量与样本参数向量进行皮尔逊相似性比较,最终确定待测样本分类。根据水灾图像识别模型,对实时监测的图像进行识别,当分割图像具有水灾纹理特征时,进行水灾报警。采集了水灾模拟实验图像,建立了图像数据库。对所提出的模型编制了相应的实验程序,进行了模型的训练和实验验证。研究了水、煤、岩石在双树复小波域泊松分布下系数统计值的分布规律,并对模型的性能进行了参数化的评估。通过实验验证了基于图像纹理特征的矿井水灾感知方法的可行性。实验表明,基于图像纹理特征的矿井水灾感知方法,识别水灾准确率大于81%。

Coal-mine flood perception method based on image texture features

Coal-mine flood is one of the most serious hazards in coal mines. The existing coal-mine flood monitoring and alarming methods have some shortcomings,such as poor adaptability,high false alarm and missed alarm rate, which are difficult to meet the needs of coal mine safe production. The authors put forward a method of coal-mine flood perception method based on image texture features:Setting cameras on the top or side of coal mine roadway and mining face supports,and collecting the real-time images of mining face and roadway floor. The texture features of water,coal and rock images are extracted by dual-tree complex wavelet transform,and the coal-mine flood image recognition model is established. The learning samples are processed by dual-tree complex wavelet transform,and the coefficients of level 1 and 2 are extracted,and the variance and expectation values are counted. The Poisson distribution model is constructed by using the corresponding variance and expectation values,and the intensity parameters of the models in all directions are estimated. After double-tree complex wavelet transform,Pearson similarity between the intensity parameter vector and the sample parameter vector of the model composed of the variance and expectation values of the first and second order coefficients is compared,and finally the classification of the samples to be tested is determined. According to the coal-mine flood image recognition model,the real-time monitoring image is recognized. When the segmentation image has the coal-mine flood texture characteristics,the flood alarm is carried out. The image of coal-mine flood simulation experiment is collected and the image database is established. The corresponding experimental program is compiled for the proposed model,and the training and experimental verification of the model are carried out. The distribution law of the coefficients statistics of water,coal and rock under the Poisson distribution in the dual-tree complex wavelet domain is studied,and the performance of the model is evaluated by parameterization. The feasibility of coalmine flood perception method based on image texture features is verified by experiments. Experiments show that the accuracy of mine flood recognition based on image texture features is more than 81%.