Research on water meter reading recognition based on deep learning
Yue Liang,Yiqi Liao,Shaobo Li,Wenjuan Wu,Taorong Qiu,Weiping Zhang
DOI: https://doi.org/10.1038/s41598-022-17255-3
IF: 4.6
2022-07-29
Scientific Reports
Abstract:At present, there are still many old-fashioned water meters in the society, and the water department needs to send staff to read the water meter after arriving at the scene with a handheld all-in-one machine. However, there are many problems in this manual meter reading method. First, a large number of meter reading work leads to low efficiency of the entire water department, consuming a lot of time and energy, and high labor costs; second, the water meters in natural scenes have problems such as serious dial contamination and other environmental factors that interfere with the meter reading staff, and the results of the meter reader cannot be verified later. In response to these problems, this paper studies a deep learning method for automatic detection and recognition of water meter readings. This paper first introduces the existing in-depth learning models, such as Faster R-CNN, SSD, and YOLOv3. Then two datasets are sorted out, one is the original water table picture dataset, and the other is a dataset cut out from the water meter image with the black bounding box showing the water meter readings. Then two plans are proposed, one is the original water table image dataset, and the other is a dataset cut out from the water meter image with the black bounding box showing the water meter readings. Finally, by comparing the three models from different angles, it is determined that YOLOv3 in the second solution has the best recognition effect, and the accuracy rate reaches 90.61%, which can greatly improve work efficiency, save labor costs, and assist auditors in reviewing the read water meter readings.
multidisciplinary sciences