基于平行图像与深度学习的绝缘子目标检测法

绝缘子是输电线路中最易发生故障的部件之一,需及时从大量图像中检测出绝缘子,为检修提供可靠依据。为解决传统方法和深度学习各自存在的局限性,提出一种基于平行图像与深度学习的绝缘子检测方法。首先,建立人工绝缘子图像数据集,并对待测的真实绝缘子图像进行颜色预选处理;然后,构建卷积神经网络进行特征提取和分类,利用正常训练和迁移学习两种方法,并采用消融实验对不同模型进行性能的分析与对比;最终,实现绝缘子目标的检测及模型指标的评价。实验结果表明,颜色预选和平行图像方法均能有效提升模型效果,使得模型的loss能更快速的收敛,分类准确率也有明显的提高,上述方法能够有效检测出图像中的绝缘子,为后期绝缘子故障的检测和处理提供了研究基础。

Insulator Object Detection Method Based on Parallel Imaging and Deep Learning

Insulators are the most vulnerable parts of transmission lines.It is necessary to detect insulators timely in a large number of images to provide a reliable basis for maintenance.In order to solve the limitations of traditional methods and deep learning methods,an insulator detection method based on parallel imaging and deep learning is proposed.Firstly,an artificial insulator image data set was established,and a color pre-selection process was performed on the real insulator image waiting to be detected.Then a convolutional neural network for feature extraction and classification was constructed.Two different schemes were planed about normal training and transfer learning to analyze and compare the performance of different models by ablation experiments.At last,target detection and index evaluation were carried out.The experimental results show that both the color pre-selection and parallel imaging methods can effectively improve the model effect,so that the loss of the model can be more quickly converged,and the classification accuracy is also improved,which provides a research basis for the detection and processing of insulator faults in later stages.