Implementation of the CNN Algorithm for Classification of Electric Meter Recording Images

Electricity is a basic human need in carrying out every activity. Electricity usage can be measured using a KWH meter. In Indonesia, electricity customers are divided into two, namely Prepaid and Postpaid. Postpaid electricity customers need to record the numbers written on the KWH meter to find out the rupiah electricity bill that must be charged. In the implementation of recording these numbers, it is not uncommon to encounter obstacles such as a locked customer's fence and so on, so the meter reading results must be re-validated. This study aims to validate the meter record images which are classified into three classes using the Convolutional Neural Network (CNN) algorithm with the Resnet34 architecture. In this study, the highest accuracy level for the Resnet34 architecture was obtained, with an accuracy rate of 97.50%.