| 引用本文: | 孙福明,高 严,许 蕊,李明渊,魏晓鸣.基于贝塞尔曲线网络的电表用电信息识别算法[J].电力系统保护与控制,2022,50(14):133-141.[点击复制] |
| SUN Fuming,GAO Yan,XU Rui,LI Mingyuan,WEI Xiaoming.Recognition algorithm of electricity consumption information of an electric energy meter based on an adaptive Bezier curve network[J].Power System Protection and Control,2022,50(14):133-141[点击复制] |
|
| 摘要: |
| 利用计算机视觉技术来快速、准确地获得用户的用电信息对于电力部门具有重要意义。针对传统算法精度低、速度慢等问题，提出了一种基于自适应贝塞尔曲线网络的用电信息识别算法。该框架集检测、识别于一体，实现了端到端的文本定位和预测。检测端结合了特征金字塔网络和残差网络，对输入图像进行特征提取，并通过四个控制点生成贝塞尔曲线，能更好地拟合文本框。识别端采用了卷积循环神经网络，引入门控循环单元替代长短期记忆网络，再结合注意力机制对文本进行识别。最后在数据集上开展五组消融实验，进行性能对比和评估分析。实验结果显示，该算法识别精度高达99.08%，且推理速度快，可被用于用电信息检测与识别的实际应用中。 |
| 关键词: 机器视觉 自适应贝塞尔曲线网络 电表 端到端 注意力机制 |
| DOI：DOI: 10.19783/j.cnki.pspc.211176 |
| 投稿时间：2021-08-28修订日期：2022-01-26 |
| 基金项目:国家自然科学基金项目资助(61976042)；兴辽英才计划项目资助(XLYC2007023) |
|
| Recognition algorithm of electricity consumption information of an electric energy meter based on an adaptive Bezier curve network |
| SUN Fuming,GAO Yan,XU Rui,LI Mingyuan,WEI Xiaoming |
| (1. School of Information and Communication Engineering, Dalian Minzu University, Dalian 116600, China;
2. Network Center, Dalian Minzu University, Dalian 116600, China) |
| Abstract: |
| The power sector can collect user electricity information quickly and accurately using computer vision technology. Given the problems of low precision and slow speed of traditional algorithms, a recognition algorithm of electricity consumption information based on an adaptive Bezier curve network is proposed. The framework integrates detection and recognition, and realizes end-to-end text location and prediction. The detection end extracts the feature from the input image by combining a feature pyramid and residual networks, and generates a Bezier curve through four control points, which can better fit the text box. A convolutional recurrent neural network is adopted at the recognition end, and the gate recurrent unit is introduced to replace the long short-term memory network, and is then combined with the attention mechanism to recognize the text. Finally, five ablation experiments are carried out on the data set for performance comparison and evaluation analysis. The results show that the recognition accuracy of the algorithm is up to 99.08%, and the reasoning speed is fast. It can be used in the practical application of electricity consumption information detection and recognition.
This work is supported by the National Natural Science Foundation of China (No. 61976042). |
| Key words: machine vision adaptive Bezier curve network electric energy meter end-to-end attention mechanism |
