Research On Intelligent Reading Algorithm Of Substation Pointer Meter Based On Deep Learning

In order to ensure the safe and stable operation of substations,the readings of these instruments need to be monitored at all times so that abnormal conditions can be detected and dealt with in a timely manner.Since most of the traditional pointer-type meters in substations lack communication interfaces,they need to be read manually,and manual reading has problems such as low efficiency,high cost and dangerous working environment,so it is urgent to study an intelligent reading algorithm for pointer-type meters in substations.The pointer meter reading process includes three stages,such as dial area detection,low-quality dial image enhancement and reading recognition based on feature information,etc.This paper proposes an intelligent reading algorithm for substation pointer meters by combining deep learning and digital image processing technology to address the problems in the pointer meter reading process.Aiming at the problems of traditional graphic detection algorithms in detecting dial regions that are susceptible to complex background interference and low robustness,the YOLOv3 target detection algorithm based on deep learning is proposed for the dial region localization task.The multi-scale feature detection structure of YOLOv3 is improved by analyzing the distribution characteristics of the dial region in the image,so that it can accurately detect the meter targets at different scales in the image,and on this basis,the ATSS algorithm is further introduced to improve the training strategy and use focal loss to improve the negative sample confidence loss to further improve the detection accuracy.To address the problem that the dial feature information cannot be accurately extracted from the dial area affected by dark light,the image enhancement algorithm based on deep learning is proposed to recover the dark-light dial image,and since deep learning is mostly data-driven and there is no paired dataset for the image enhancement task of pointer-type meters,the unsupervised learning-based Ex CNet algorithm is proposed to perform the dark-light meter image enhancement task using the Ex CNet algorithm based on unsupervised learning.In order to preserve and enhance the detail-rich areas such as the dial internal scale and hands,a bootstrap filter is introduced to decompose the image,and the CLAHE and BM3D algorithms are introduced to enhance and suppress the noise in the decomposed detail component image,resulting in an enhanced dial image with more obvious detail information.For the angle method reading is difficult to effectively calculate the pointer rotation angle and other problems,proposed the use of rotation jamming algorithm to obtain the minimum external rectangle of the pointer area,based on the position relationship between the center of mass of the rectangle and the center of mass of the pointer to determine the pointer tip direction,combined with the slope of the fitted straight line to determine the pointer rotation angle,and then calculate the meter reading.The experimental results show that the absolute error of the intelligent reading algorithm of the pointer meter proposed in this paper is less than 0.4,and the quoted error is less than 2%,which can meet the accuracy requirements of the pointer meter reading in the substation.