基于多尺度密集对比增强自监督的简统化接触网缺陷检测

【摘要】 针对简统化接触网缺陷检测缺陷样本少、目标小、类别多的问题，提出基于多尺度密集对比增强自监督的简统化接触网缺陷检测方法。利用编码器提取多尺度特征，构建多尺度对比和密集对比更新模型，通过迁移学习的方式获取预训练模型，进而获得高精度缺陷检测模型。实验证明本文所述方法具有一定的可行性和有效性，对于研发可实际应用的简统化接触网缺陷检测系统具有重要意义。更多还原

【Abstract】 With regard to the problems of few samples, minor targets and many categories of defects of simplified and unified OCL for inspection. The paper proposes a simplified and unified OCL defect inspection method based on multi-scale dense contrast enhanced self-monitoring. The encoder is used to extract multi-scale features, build multi-scale contrast and intensive contrast update models, and obtain pre-training models through transfer learning, thus obtaining high-precision defect inspection models. The experiment proves that the method described in the paper is feasible and effective, and it is of great significance for the development of a practical simplified and unified OCL defect inspection system.更多还原