【摘要】 针对铁路变电所辅助监控系统中图像识别技术面临的样本不均衡、复杂环境干扰及实时性不足等问题，提出一种基于正负样本融合的轻量化图像识别方法。通过引入动态加权对比学习策略，结合MobileNetV3轻量化网络与多尺度特征融合机制，显著提升模型故障检测精度与鲁棒性。经试验及实际实用表明：采用正负样本融合技术，模型故障识别准确率达93.5%以上，误报率降低至3.1%以内；本文所提方法在接触网异物检测、变压器漏油预警等场景中具有较好的应用效果，为铁路安全运维提供了可靠技术支撑。更多还原

【Abstract】 With regard to the problems of sample imbalance, complex environmental disturbance and insufficient real-time performance faced by image recognition technology in auxiliary monitoring system of railway traction, a lightweight image recognition method based on positive-negative sample fusion is put forward. By introducing a dynamic weighting contrastive learning strategy, and combing the MobileNetV3 lightweight network and multi-scale feature fusion mechanism, the fault detection accuracy and robustness of the model are significantly improved. Experiments and actual applications show that fault recognition accuracy of the model is more than 93.5% and the false alarm rate of is reduced to less than 3.1% by adopting positive-negative fusion technology. The method proposed in the paper has better application effect in these scenarios such as foreign object detection of OCS and transformer oil leakage early warning, and providing reliable technical support for railway safety operation and maintenance.更多还原