Research on Recognition of Pointer Meter Based on Improved East Algorithm

Abstract

In order to solve the problems of low recognition accuracy and large memory occupied in the algorithm of pointer meter identification in substation, this paper proposes an improved algorithm of pointer meter automatic identification based on EAST algorithm, Tiny East. Firstly, the algorithm uses the lightweight neural network MOGA to replace the backbone network of East algorithm, thus reducing the memory occupied by the algorithm; Then, the feature pyramid is added to backbone network to strengthen the feature extraction. Besides, we use NMS local sensing algorithm to screen the reading box to get the scale number and position of the instrument. Thus, positioning the pointer line and the center of the instrument; Finally, the identification of the instrument is completed by combining the identified digital and pointer angles. Experimental results show that compared with the unimproved EAST algorithm, the number of parameters of the proposed TINY-EAST algorithm is greatly reduced, and the average recognition accuracy is higher than 98.5%. Moreover, the proposed TINY-EAST algorithm has good accuracy and stability in the complex background of pointer instrument and can meet the application requirements of practical substation.