Recently, with the expansion of the demand response market, the need to automatically read power meters is increasing. In addition, computer vision using deep neural networks has rapidly developed, and object detection, the task of finding a specific object in an image, has now reached high accuracy. There are several studies on the subject of automatic power meter reading. However, studies that apply object detection to reading meters are difficult to find. Automatic power meter reading is generally divided into the task of finding the number corresponding to the usage and the task of recognizing the number. In this paper, we only deal with finding numbers using object detection. Training is needed to create an object detection model that finds only the numeric area in the meter. The important factors in training a model are the amount of training data and the number of training epochs. Training a lot of epochs with a lot of training data will show high detection performance, but it takes a lot of time to prepare for training and training data. In this paper, we present the amount of training data and the number of epochs that can take the least time to train an object detection model with a detection performance of over 99%. Also, the results of various experiments performed to find this parameter are recorded in the paper. 1
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.