Analog Digit Electricity Meter Recognition Using Faster R-CNN

Abstract

The current measurement of electricity consumption use a device called kWh meter that logs the total consumption of electricity, unfortunately to record the data to the electricity provider in Indonesia the employee of the provider still need to come and check the eletrical usage manually. In this paper we created a Deep Learning model based on Faster R-CNN to reads the digit from an analog electricity meter using dataset from the UFPR-AMR Dataset From the training we achieved the best model with the configurations of 90:10 for the data partition split, batch size of 3, learning rate of 0.04, and epoch of 7000 and gained results with accuracy of 99.67%, recall of 98.04%, and precision of 98.04%