Research and Implementation of Pointer Instrument Reading Recognition Method for Power Plant Inspection Robot

Instrument recognition is the core task of the equipment inspection subsystem, where the robot equipped with a camera sensor captures images of pointer-style instruments and returns them for further processing. This paper investigates the practical problem of accurately processing instrument images to obtain target instrument dial data. We propose a method for pointer-style instrument reading recognition based on the combination of YOLOv5 and U-2-Net. The YOLOv5 model is used to train and perform object detection to extract and locate the instrument dial region. The U-2-Net model is employed to obtain the scale and pointer positions within the dial, and the reading is calculated based on their relative positions. The whole instrument identification process is communicated to the outside world through the ROS message mechanism and Web API interface. Through this method, the equipment inspection subsystem can accurately solve the pointer dashboard reading problem, can greatly improve inspection efficiency, and can meet the daily inspection work of the environmental robot in the power plant.