Key Feature Recognition Algorithm for Energy Meters under Convolutional Neural Networks

During the inspection process, a large number of massive images of power equipment would be generated. Therefore, the first problem to be solved is how to extract effective data for acceptance and operation management of power equipment more accurately and efficiently. The research purpose of this article is to use Convolutional Neural Networks (CNN) and image recognition technology, and on the basis of investigating and analyzing the current application status of energy meter processing processes. It provides a new targeted image recognition scheme for electricity meters, which can achieve the recognition of electricity meter photography, barcode and reading, as well as the verification and storage of electricity meter information. This article conducts color image grayscale processing, image enhancement, binarization processing, image recognition area positioning, and selection of light sources for electric energy meters. The image recognition area positioning includes the positioning of the electricity meter barcode area and the positioning of the electricity meter barcode area. The recognition rate of the CNN for the entire instrument can reach 98.5%. Among them, the barcode recognition rate can reach 97.8%, and the recognition rate of the bottom reading of the electricity meter can reach 99.2%. This article helps to reduce errors caused by human factors and is of great significance for achieving the intelligence of electricity meters.