基于改进U-Net的输电线路绝缘子图像分割方法研究

Image Segmentation Method of Transmission Line Insulator Based on Improved U-Net

针对输电线路巡检航拍的绝缘子图像存在背景复杂、对比度不明显、图像质量不能保证等情况造成绝缘子分割精度不高的问题,提出一种基于注意力模型改进U-Net网络的分割方法。首先以VGG16替换主干特征提取网络,增强网络的适用性;同时在下采样过程中引入注意力模型,增强对绝缘子目标的辨识能力,抑制背景、噪声等干扰信息,实现更加精确的分割。实验结果表明:CBAM注意力模型与U-Net网络相结合的方式效果最好,平均重叠度可达96.57%。

Targeting the low segmentation accuracy of insulators due to the complex background,inconspicuous contrast and poor image quality of insulator images captured by the aerial photography of transmission line inspections,the paper propose a segmentation method based on improved U-Net network with attention model.Firstly,VGG16 instead of backbone feature extraction network is used to enhance the applicability of the network.Then an attention model is introduced in the down-sampling process to enhance the ability to identify the insulator targets,and to suppress the interference information such as background and noise,achieving more accurate segmentation.The experimental results show that the combination of the CBAM attention model and the U-Net network works best with an average overlap of 96.57%.