Research On Video Based Recognition Algorithm For Violation Behavior Of Well Site Workers

With the continuous improvement of the safety demand of oilfield well sites,a large number of monitoring equipment is deployed in well sites for safety monitoring work,but manual long-term monitoring is required,and there are problems of high work intensity and low feedback efficiency.In recent years,video behavior recognition has become one of the research hotspots in the field of computer vision,and has been widely used in public security border defense,intelligent transportation and other fields.The traditional behavior recognition method is directly applied to the real oilfield well site environment,and there are many training parameters and the recognition accuracy is not high due to the interference of background factors.Therefore,this paper uses deep learning algorithms to construct a video-based behavior recognition model and improve it,so as to accurately and quickly identify the violations of workers in the well site and ensure the safety of the well site in the oil field.The main research contents of this paper are as follows:1.Propose an improved video behavior recognition model based on attention mechanism.The problem of low recognition accuracy is caused by the interference of relatively static background information such as railings,pumping units,trees,and other noisy backgrounds in the video.Firstly,this paper uses the attention mechanism CBAM to improve the 3D residual network Res Ne Xt,enhancing the extraction of significant behavioural features from recognition models based on channel and spatial dimensions,reducing background noise interference and improving the recognition accuracy of the model.Secondly,a dataset of well site workers’ violations was constructed,containing smoking,no＿helmets,climbing pumping machines and answering phone calls.Through experiments on public datasets and the constructed well site violation dataset,it is demonstrate that the improved recognition model using the attention mechanism effectively enhances the extraction of key behavior feature information,improves recognition accuracy and has strong generalization ability.2.Propose a video behavior recognition model based on depthwise separable convolution and GRU optimization.Firstly,the improved 3D residual network Res Ne Xt model has the problems of large number of parameters and long training period.In this paper,the residual network model is lightly improved by using 3D depthwise separable convolution to perform separation operations on 3D convolution in the residual module.In addition,this method can effectively reduce network parameters while ensuring that the deep network can extract multiscale feature information,so that more detailed information can be preserved.Secondly,a gated recurrent neural network GRU is added to the feature extraction backbone network for optimization,enhancing the model’s extraction of long time sequence features,improving the model’s focus on key video frame time information,and further improving the robustness of the model’s feature extraction.Finally,use the Log(softmax)classification function to obtain a more stable result classification output.The experimental results show that the recognition model based on depthwise separable convolution and GRU not only improves recognition accuracy and enhances feature extraction ability,but also effectively reduces recognition model parameters and improves training speed.Experiments have shown that the improved deep network recognition network model proposed in this paper can effectively improve the recognition accuracy,enhance the effective extraction of key features from the model and reduce the network parameters.