Research On Instrument Identification Technology Of Substation Based On The Deep Learning

The substation plays a vital role in the stable operation of the power system.Therefore,the staff needs to monitor the meter readings of the substation to detect abnormal conditions in time.In the past,the reading instrument mainly adopts the manual reading method which has high cost,low efficiency,high recognition error rate and so on.It is urgent to research a set of intelligent recognition meter algorithms used in substations.This paper studies the algorithms of meter image preprocessing,meter frame detection,and meter reading recognition systematically.This paper briefly introduces the split-combination text meter recognition and indication mark color recognition,and designs the software test module to display the recognition results of the meters.The primary research of this paper:(1)We collect data images of ‘fen’ and ‘he’ text instruments,fault indication signs and digital instruments,and label the self-made instrument image data set.This paper uses the digital image processing technology to pre-process the meter dial image and the digital display area of the digital meter.When the meter dial area is preprocessed,the image adaptive enhancement operation is performed first to enhance the image brightness.Secondly the image is Gaussian blurred.Then the image is grayed out to reduce the color image dimension,improve the model training speed and reduce the information space.Finally,the initial image and the image data are rotated,mirrored and translated.(2)When the digital meter digital display areas are processed,firstly the digital display areas are tilt corrected.Secondly the corrected images are grayed out.Then the images grayed out are binarized.Finally the binarized images are expanded and corrupted by morphological algorithm.(3)We recognize ’fen’ and ’he’ meters and fault indication.Firstly the migration learning is used to select a suitable feature extraction network for the Faster R-CNN algorithm.Secondly the Faster R-CNN algorithm is applied to the recognize ’fen’ and’he’ meters and fault indication signs of substation.At the same time to highlight the superiority of the Faster R-CNN algorithm,the SSD algorithm is used to compare with the Faster R-CNN in terms of the detection accuracy and recognition results.Finally the results are displayed in the software interface.(4)We recognize digital meters.After the Faster R-CNN algorithm and the SSD algorithm are applied to the recognize meter dials respectively,the Faster R-CNN algorithm can be concluded to extract the meter bezel accurately and the detection results are not affected by the ambient light.After the meter frame areas are extracted,a meter dial digital display area recognition based on the LSTM algorithm and the CTC algorithm is proposed.The recognize result is displayed in the software interface system.