基于特征迁移的螺栓图像超分辨率处理方法

PDF下载HTML

Title:

A method for super resolution processing of bolt image based on feature transfer

作者:

戚银城^1,2, 耿劭锋¹, 赵振兵^1,2, 吕雪纯¹, 孙梦¹

1. 华北电力大学 电子与通信工程系，河北 保定 071003;
2. 华北电力大学 河北省电力物联网技术重点实验室，河北 保定 071003

Author(s):

QI Yincheng^1,2, GENG Shaofeng¹, ZHAO Zhenbing^1,2, LYU Xuechun¹, SUN Meng¹

1. Department of Electronic and Communication Engineering, North China Electric Power University, Baoding 071003, China;
2. Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding 071003, China

关键词:

螺栓; 图像超分辨率; 深度学习; 特征迁移; 输电线路; 缺陷检测; 神经网络; 智能巡检

Keywords:

bolt; image super resolution; deep learning; feature transfer; transmission line; defect recognition; neural network; ntelligent patrol

分类号:

TP391; TM726

DOI:

10.11992/tis.202201009

摘要:

针对输电线路巡检采集的螺栓图像存在模糊、分辨率低等问题，本文根据螺栓之间相似度较高的特点，提出了一种基于特征迁移的螺栓图像超分辨率处理方法。本文首次将特征迁移引入螺栓图像超分中，先对比低分辨率图像与清晰参考图像的特征区域，将图像之间相似度高的区域进行特征迁移，并根据迁移特征的相似度调整迁移特征的比例，然后在感知损失函数中加入迁移特征相似度的约束，保证迁移特征的准确性。不同超分模型的螺栓图像超分实验结果表明，本方法的螺栓超分图像更清晰，峰值信噪比、结构相似性指标更优；此外，超分前后螺栓图像的缺陷识别实验对比结果表明，超分后螺栓的缺陷识别准确率提升了3.61%，实验结果验证了本文方法的有效性。

Abstract:

Aiming at the problems of blurring and low resolution of bolt images collected during inspection of transmission lines, a super-resolution processing method of bolt images based on feature migration is proposed according to the characteristics of high similarity between bolts. In this paper, feature transfer is introduced into bolt image super resolution for the first time. First, compare feature regions of the low-resolution image and the clear reference image, then carry out feature transfer of the regions with high similarity between the images, and then adjust the proportion of transfer features according to similarity of the transferred features. Then, the constraint of similarity of the transfer features is added to the perceptual loss function to ensure accuracy of the transfer features. The results of the bolt image over-segmentation experiments with different over-segmentation models show that the bolt over-segmentation images of this method are clearer, and the PSNR and SSIM indicators are better; The defect recognition accuracy rate of the bolt increased by 3.61% after ove-segmentation. The experimental results fully verify effectiveness of this method in solving the problem of blurred bolt images.