Abstract: The phenomenon of oil theft at well sites is an important issue that affects the safe production and stable operation of oil fields. The current behavior recognition methods pay less attention to the need for detecting oil theft in well pads, and there are often limitations in the application of the oil theft target feature recognition process. An algorithm for identifying oil theft behavior at well sites is proposed based on 3D attention residuals. This network consists of multiple three-dimensional attention residual blocks, which embed channels and spatiotemporal attention modules in each residual block to obtain more feature discrimination information and enhance the model’s attention to important features. The effectiveness of the algorithm is varified on the dataset of oil theft behavior at the well site. The experimental results indicate that, compared to similar algorithms, this method has higher recognition accuracy. It can provide a reference for the practical application of automatic detection of oil theft behavior in oilfield well sites.

Key words: oilfield theft, 3D convolution, behavior recognition, residual module, attention mechanism