Safe production is a top priority for the state and coal mining enterprises.However,the quality of water exploration and release operations cannot be guaranteed,which leads to frequent flooding disasters,which has a huge impact on the economy,life and property.So far,relevant enterprises have generally adopted manual inspection and acceptance of water exploration operations to prevent water penetration accidents.However,manual inspection and acceptance has problems such as long time consumption,low efficiency,and general paper-based storage that is not easy for long-term storage.After studying and analyzing the working process of water exploration operations,it is proposed to use computer technology to recognize water exploration operations,so that water exploration operations can be intelligently accepted.In this way,it plays a monitoring role in the implementation process of water exploration operations,and further improves the prevention and control level of coal mine flooding disasters.In this paper,a three-dimensional convolutional neural network is used to recognize the action of water exploration,and an improved 3DCNN model is established to solve the problem of insufficient feature extraction capabilities of traditional convolutional neural networks.First,use the Soft Max cross-entropy loss function behind the convolutional layer to discriminate and learn the sample feature information to capture deeper feature information to obtain a richer network feature map;secondly,combine the batch regularization operation to regularize the pheromone points Redistribute,speed up the network convergence speed,and improve the generalization ability and robustness of the model.Finally,the study found that in the process of convolution,it is very easy for neurons to be inhibited and lose their activity directly.The ReLU function of the excitation layer is upgraded and optimized to the Leaky ReLU function,and silent neurons are restored and used during the non-linearization process.Let more neurons participate in learning,so as to extract richer information more efficiently.The experimental results show that compared with the existing methods,the training efficiency and recognition accuracy of the proposed method on the UCF-50 public data set and self-made water exploration operation data set are significantly improved,which can verify the effectiveness of the model.This method can provide a new technical reference for the field of intelligent inspection and acceptance of water exploration operations,and has certain significance for engineering practice. |