| With the rapid development of China’s oil and gas enterprises’ production towards intelligent transformation,intelligent inspection of oil and gas field operation stations such as natural gas gathering stations,gas distribution stations,and natural gas power plants is now an important measure for oil and gas field enterprises.Implementing intelligent detection of instruments in operation stations can reduce the working pressure of manual inspection and improve the efficiency of intelligent inspection.This article applies deep learning technology to intelligent detection of instruments.For pointer and level gauge instruments in oil and gas field production areas,a detection algorithm for core components of oil and gas field instruments based on an improved semantic segmentation model is proposed.An improved snake optimized pointer table detection algorithm based on key point detection and an improved level gauge detection algorithm based on FCN regression model are designed,which has certain guiding significance for oil and gas field instrument detection.The main work and research content are as follows:(1)Due to the problems of messy image background and multiple types of instruments in oil and gas fields,a detection algorithm for the core components of oil and gas field instruments based on the improved semantic segmentation model was proposed.By integrating codecs and multi branch structures,and introducing self attention mechanism,an improved semantic segmentation model for the core components of oil and gas field instruments has been constructed,which improves the detection accuracy and speed of field instrument images with cluttered backgrounds.The effectiveness of the algorithm was verified through simulation experiments.(2)Due to the problems including perspective distortion,damp liquid refraction and occlusion of field pointer tables such as temperature meters and pressure gauges,an improved Snake optimum pointer table detection algorithm based on key points was proposed.A pointer table key point detection model was built,and through semantic segmentation and key point detection methods,pointer table readings were obtained under the occlusion of the dial.An improved snake optimization algorithm was used to correct pointer table readings with image distortion.The feasibility of this algorithm was verified through simulation experiments.(3)Due to the problems including slim shape of the field image,the long shooting distance and the small proportion of the image area in the liquid level meter installed in oil and gas field facilities such as desulphurization towers,regeneration towers,gas tanks,etc.,an improved liquid level meter detection algorithm based on FCN regression model was proposed.By analyzing the geometric features of the image,a horizontal convolutional layer,horizontal dimensionality reduction,and regression detection head mechanism were introduced,and a FCN regression detection model for the liquid level gauge was established to improve the robustness of the algorithm.Simulation experiments show that this algorithm can improve the detection accuracy of on-site liquid level gauges in oil and gas fields. |
