【摘要】 为准确识别铁路接触网巡检图像与几何参数测量数据中的关键特征（如定位线夹、吊弦线夹），首先利用图像特征、采用改进的YOLOv3检测模型进行测试，发现该类方法易受线路上相似物件的干扰，误判率高；又依据接触网“之”字形设计和弹性悬挂特性，提出数据趋势识别法，该方法在数据平稳时效果良好，但易受接触网非常规布局或数据波动的影响导致漏判。为综合二者优势、克服单一方法缺陷，进一步提出利用D-S证据理论，对上述2种方法的识别结果进行决策级融合，使吊弦和定位器线夹识别的F1值优于95%，识别效果优于单一方法，能更有效、更可靠地识别接触网关键特征，为进一步保障供电安全提供技术支撑。更多还原

【Abstract】 To accurately identify key features in railway OCS inspection images and geometric parameter measurement data(such as steady ear and dropper clip), this paper tests an improved YOLOv3 detection model using image features. It is found that this method is prone to interference from similar objects on the line, resulting in a high misjudgment rate. Additionally, based on the zigzag design and elastic suspension characteristics of the OCS, a data trend recognition method is proposed. This method performs well when the data is stable but is susceptible to missing judgments due to unconventional layouts of the OCS or data fluctuations. To integrate the advantages of both methods while overcoming their individual limitations, this paper further proposes a decision-level fusion strategy based on D-S evidence theory, combining the recognition results from the two aforementioned methods. The proposed fusion strategy achieves an F1 value exceeding 95% for identifying dropper clips and steady ears, which outperforms either of the single methods. This enables more effective and reliable detection of critical OCS features, providing technical support for enhanced power supply security.更多还原