基于YOLO-E与改进OCRNet图像分割的变电站仪表读数自适应识别方法

Substation Meter Readings and Dial Information Identification Method Based on YOLO-Eand Enhanced OCRNet Image Segmentation

变电站仪表读数的准确识别对实时感知设备运行状态,提高变电设备运维智能化水平意义重大。然而,现有基于指针偏转角度检测的变电站仪表读数识别方案准确性差,未能结合表盘色带判别设备状态,无法自适应表盘量程。对此,提出了一种基于YOLO-E改进OCRNet图像分割的变电站仪表读数识别方法,首先提出了基于YOLO-E的表盘位置检测算法,通过透视变换实现图像校准;其次,使用极化注意力模块分支改进OCRNet网络结构,提出基于改进OCRNet的表盘分割算法,实现表盘刻度、指针及色带的准确分割提取;最后,基于PGNet从表盘文本中自适应识别量程信息,结合指针与刻度的分割结果实现读数识别。算例证明,与其他先进电力视觉算法相比,所提方法能适应不同量程的仪表,实现了读数与设备运行状态的准确识别,有助于推动运检数字化转型。

Leveraging electric vision imaging technology to identify instrument readings in substations offers substantial benefits in real-time equipment monitoring and elevating operational and maintenance intelligence.However,many existing instrument image recognition solutions for substations rely on pointer deflection angles,despite the need for enhanced precision and robustness.These methods overlook critical data,such as device status,numerical intervals reflected by dial color,and dial character information.This paper introduces a novel method for automatic recognition of instrument readings and dial information in substations.First,it proposes a dial type position detection algorithm based on YOLO-E,achieving image calibration through perspective transformation.Second,building upon OCRNet's object region context extraction structure,it incorporates a parallel branch with polarized self-attention to rationally utilize channel feature maps with varying weights.This results in a dial segmentation algorithm based on an improved OCRNet.By segmenting scales,pointers,and color bands,this method achieves precise segmentation and identification of meter readings and crucial additional information.Finally,using PGNet,the method recognizes dial information,enabling automatic acquisition of data like meter range parameters and readings for multi-range dials.A case study demonstrates that,compared to other advanced electric vision algorithms,the proposed method not only enhances reading recognition accuracy but also effectively detects and extracts additional dial information.This advancement supports the digital transformation of operations and maintenance.