Study on Text Detection and Positioning Method of Utility Pole Identification Plate Based on Improved EAST

The utility pole identification plate contains important grid line information. At present, it is mainly recorded manually, and large quantities of repetitive work consume a giant amount of human resources. At the same time, the domestic power industry has not yet formed a unified target detection and identification engine, and no related research has been retrieved in the field of utility pole image detection. Therefore, it is imperative to use computer vision technology to replace manual and visual inspection and improve the automatic management level of the power grid. Based on this understanding, using the latest image detection model construction method, this paper adopts an improved EAST network to detect the utility pole identification plate. The feature extraction layer uses VGG-16 network to accelerate the calculation. The entire network is trained end-to-end to obtain the utility pole identification area directly. A large number of experiments were performed by constructing a pole image data set, and the text detection accuracy rate was 83.2%, which verified the effectiveness of the detection system.