一种基于智慧视觉的电力线路检测技术

A Power Line Inspection Technology Based on Intelligent Vision

摘要: 针对电力线路人工巡检效率低、传统图像处理方法泛化能力差的问题，本文提出一种基于VGG16迁移学习的故障检测方法。首先构建包含绝缘子破损图像数据集，通过数据增强扩充样本；然后利用ImageNet预训练的VGG16模型，采用冻结卷积层、微调全连接层的迁移学习策略，将通用特征提取能力迁移到电力线路故障识别任务，最后在自建数据集上进行训练和测试。实验结果表明，训练损失与验证损失曲线保持同步下降趋势，无明显过拟合现象。该方法对复杂背景下的故障目标具有较好的识别能力，验证了迁移学习在小样本电力故障检测中的有效性。

Abstract: Aiming at the problems of low efficiency in manual inspection of power lines and poor generalization ability of traditional image processing methods, this paper proposes a fault detection method based on VGG16 transfer learning. Firstly, a dataset containing insulator damage images is constructed, and the samples are expanded through data augmentation. Then, the pre-trained VGG16 model on ImageNet is utilized, and a transfer learning strategy of freezing the convolutional layers and fine-tuning the fully connected layers is adopted to transfer the general feature extraction ability to the power line fault recognition task. Finally, training and testing are conducted on the self-built dataset. The experimental results show that the training loss and validation loss curves maintain a synchronous downward trend, with no obvious overfitting phenomenon. This method has a good recognition ability for fault targets in complex backgrounds, verifying the effectiveness of transfer learning in small-sample power fault detection.