Automatic Reading Method for Multi-Class Pointer Meters in Power Applications

作者：Shiyun Chen, Jing Cao, Jingeng Li, Min Guo, Wenbo Xiang · 年份：2024 · DOI：10.23919/ccc63176.2024.10662416 · 被引用次数：4 · 研究领域：Algorithms and Data Compression、Advanced Data Compression Techniques、Speech and Audio Processing

Existing automatic reading methods for pointer gauges mostly target at a single type of meter. However, in power applications, there are a wide variety of meters with different styles and measuring ranges. If each meter uses an unique model, the algorithm would be too complex for practical application. Therefore, this paper proposes an universal reading model for pointer instruments based on dial key point detection and range recognition. Unlike existing two-step processing frameworks that first detect the dial and then identify the readings, the model proposed in this paper directly detects three key points of the dial including the pointer endpoint, scale point, and center point, as well as the scale number area using a modified CenterNet network. Then, the scale number area is extracted and fed into an STN-CRNN network for scale recognition to obtain precise ranges. Without the need of dial classification, the model directly obtains the meter reading. In the tests on 533 substation meter images containing 41 types of gauges with different ranges, this method achieves an 86.43% recognition accuracy when the reading error is not higher than $\pm 5 \%$.