Action Recognition Of The Coal Mine Water-probing Unloading Lever Based On 3D CNN

In recent years,China's coal mine safety production situation has gradually improved,but severe accidents occurred from time to time.The proportion of water accidents in these accidents cannot be ignored,because the losses are distressing.For preventing the occurrence of water accidents,it is necessary to ensure the quality and quantity of water-probing operations.But the existing supervision methods are manual supervision,which is inefficient and wasteful of the resources.This paper proposes an action recognition model of the coal mine waterprobing unloading lever based on 3D CNN.This method can automatically complete the extraction of action features and omit the process of manually designing action features in traditional action recognition methods.For the action recognition model based on 3D CNN,the accuracy of action recognition is mainly related to the structure of the model.The extraction of action features is mainly finished by convolutional layers,and the quality of feature extraction directly determines the recognition accuracy of the model.Therefore,this paper compares the different structures action recognition models based on 3D CNN through experiments from three aspects: the size of the convolution kernel,the number of convolution kernels,and the number of convolutional layers.At the same time,experiments are designed to verify the effect of adding batch normalization layer on the accuracy of the model.The experimental results show that the optimal convolution kernel size based on the 3D CNN action recognition model is 3 × 3 × 3;the increase in the number of convolution kernels and the number of convolution layers will increase the accuracy of the model,but also increase the parameters of the model;adding a batch normalization layer will also improve the accuracy of the model and make the accuracy of the model more stable.For the action recognition of the unloading lever,the method proposed in this paper first samples the surveillance video of the water-probing,the fixed number of pictures are extracted from the video using Open CV while reducing its resolution.Secondly,the spatial and temporal information in the video is extracted simultaneously through the three-dimensional convolution feature extraction networks.Thirdly,the pooling layer is adopted to reduce dimensionality.Finally,unloading lever action has been classified by the Softmax classifier.The experiment compares the accuracy of the model with different sampling frames,different resolutions,and different learning rates,and by adding the batch normalization layer the convergence speed and accuracy of the model has increased.The recognition effect of action recognition model of the coal mine water-probing unloading lever based on 3D CNN is verified by the experiments on our own dataset,the results of which show that a higher recognition rate in action recognition has been gained.On our own data set,the accuracy rate has been improved up to 98.86% with a 10% increase through adding batch normalization layer to the 3D CNN action recognition model.