TY - GEN
T1 - Deep Learning and Optical Character Recognition for Digitization of Meter Reading
AU - Chong, Yue Jiet
AU - Huat Chua, Kein
AU - Babrdel, Mohammad
AU - Hau, Lee Cheun
AU - Wang, Li
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - One of the important steps in digital transformation is to make all the instrumental devices connected to the cloud. However, replacing the existing analogue meters with the cloud-connected digital meters can be very costly especially for industrial grade meters. In this project, a deep learning model based on Single Shot Detector (SSD) MobileNet V2 and an optical character recognition (Tesseract OCR) engine are developed for the low-cost digitization of analogue meter readings. The deep learning model is trained with a dataset of 750 meters' images, and it is used to detect the region of interest where the meter's readings are located. The OCR is used to convert the readings to string datatype. The image processing techniques via the OpenCV library are adopted for enhancing the quality of the region of interest (ROI). The model is developed in Python and the evaluation is carried for various types of meters, illumination conditions, and backgrounds. The results show that the deep learning model and OCR accuracies are 95% and 93%, respectively.
AB - One of the important steps in digital transformation is to make all the instrumental devices connected to the cloud. However, replacing the existing analogue meters with the cloud-connected digital meters can be very costly especially for industrial grade meters. In this project, a deep learning model based on Single Shot Detector (SSD) MobileNet V2 and an optical character recognition (Tesseract OCR) engine are developed for the low-cost digitization of analogue meter readings. The deep learning model is trained with a dataset of 750 meters' images, and it is used to detect the region of interest where the meter's readings are located. The OCR is used to convert the readings to string datatype. The image processing techniques via the OpenCV library are adopted for enhancing the quality of the region of interest (ROI). The model is developed in Python and the evaluation is carried for various types of meters, illumination conditions, and backgrounds. The results show that the deep learning model and OCR accuracies are 95% and 93%, respectively.
UR - http://www.scopus.com/inward/record.url?scp=85133469422&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85133469422&partnerID=8YFLogxK
U2 - 10.1109/ISCAIE54458.2022.9794463
DO - 10.1109/ISCAIE54458.2022.9794463
M3 - Conference contribution
AN - SCOPUS:85133469422
T3 - 2022 12th IEEE Symposium on Computer Applications and Industrial Electronics, ISCAIE 2022
SP - 7
EP - 12
BT - 2022 12th IEEE Symposium on Computer Applications and Industrial Electronics, ISCAIE 2022
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 12th IEEE Symposium on Computer Applications and Industrial Electronics, ISCAIE 2022
Y2 - 21 May 2022 through 22 May 2022
ER -