基于双级校正的输电线缆型号识别

Transmission Cable Model Identification Based on Dual-Stage Correction

中文关键词:  输电线缆型号识别  双级校正  区域级分割  注意力机制

英文关键词:transmission cable model identification  dual-stage correction  regional segmentation  attention mechanism

中文摘要:

      识别并准确读取输电线缆型号是电力系统设备管理和运行维护的关键任务。针对现有方法在弯曲形态复杂、背景干扰强的输电线缆场景中存在检测精度低、速度慢的问题，提出了一种基于双级校正的输电线缆型号识别方法。该方法首先利用区域级分割技术精确提取文本区域位置；其次，设计双级校正策略，采用薄板样条（Thin Plate Spline,TPS）插值实现粗校正，并通过偏移映射完成精细校正，从而优化文本的几何分布；最后，在识别模块中融入基于注意力的特征优化机制，增强了对不规则、模糊文本的识别能力。实验结果表明，所提方法在自建输电线缆数据集及公开数据集COCO-Text和CTW-1500上整体表现优异，较部分传统方法平均识别准确率提升超过2.7%。同时，在高反光及复杂背景场景中，其识别准确率达到85.3%，实现了检测效率和识别精度的有效平衡。

英文摘要:

      Identifying and accurately reading transmission cable models is a key task in power system equipment management and operation maintenance.In order to solve the problems of low detection accuracy and slow speed of existing methods in transmission cable scenes with complex bending patterns and strong background interference,a transmission cable model identification method based on dual-stage correction is proposed.Firstly,region-level segmentation is used to accurately extract the text region location.Secondly,a dual-stage correction strategy is designed,which adopts thin plate spline(TPS) interpolation to achieve coarse correction,and completes fine correction through offset mapping,so as to optimize the geometrical distribution of the text.Finally,an attention-based feature optimization mechanism is incorporated in the recognition module,which enhances the recognition ability for irregular and fuzzy text.The experimental results show that the proposed method performs well on the self-constructed transmission cable dataset and the publicly available datasets COCO-Text and CTW-1500,with an average recognition accuracy improvement of more than 2.7% over some of the traditional method.Meanwhile,its recognition accuracy reaches 85.3% in high reflection and complex background scenes,achieving an effective balance between detection efficiency and recognition accuracy.