基于轻卷积神经网络的电力电缆绝缘损伤图像检测方法

A power cable detection algorithm based on a light convolutional neutral network

电缆良好的绝缘性是保证电缆安全运行的重要保障。针对传统电力电缆绝缘损伤检测方法存在检测流程复杂和无法大规模整体检测的缺点,本文结合图像无损接触方式和深度学习方法,提出了一种基于深度学习的电力电缆图像破损批量检测方法。该方法创新性地建立了基于残差和深度可分离模块的轻深度卷积神经网络模型,和以往的卷积神经网络模型相比,网络极好地平衡了系统的识别时间和识别精度,能实现高效、无损、快速的大规模电缆外表面多样化异常检测。和传统学习方法和已有深度卷积神经网络模型的实验结果对比表明,本文方法具有良好的实时性、鲁棒性和识别率,识别正确率达到99.47%。

Power cable,as the important power instructions is a crucial carrier for power transportation. The process of traditional cable detection methods is complex,and it could not be used to detect a mass of whole cables.Therefore,a cable abnormalities detection algorithm based on image acquisition and deep learning in this paper is introduced.The novelty of this method is that a light convolutional neutral network based on depth-wise separable residual convolution module is involved and proposed to detect all kinds of cable abnormalities and deal with a mass of whole cables.In order to evaluate our algorithm, the comparative experiments with other learning methods is involved.The results confirm the accuracy, the robustness and real-time performance of the proposed method.