基于超分辨注意力机制改进的GIS内部细微缺陷X-DR图像检测方法

Improved X-DR Image Detection Method for GIS Internal Defects Based on Super-resolution Attention Mechanism

为了解决气体绝缘开关设备内部细微缺陷X-DR成像重影雾化、纹理不清晰、易造成误诊断等问题,提出一种基于超分辨率注意力机制改进的气体绝缘开关(gas insulated switchgear,GIS)内部细微缺陷X-DR图像检测算法。该方法以高效亚像素卷积神经网络为框架,引入双层计算机注意力机制SE模块,构成新颖的SE-ESPCN超分辨率网络。通过对GIS设备X-DR图像通道重要程度进行评判,赋予图像卷积层不同的权重,以增强图像故障细节的成像效果。实验结果表明:SE模块与亚像素卷积神经网络的融合,不仅保障了GIS设备故障区域成像的实时性,而且算法输出的高分辨率X-DR图像缺陷细节清晰可见,便于观察,提高了工作人员对缺陷检测的效率与准确性,为实际工程中GIS设备X-DR成像系统改良提供了一定的参考。

In order to solve the problems such as X-DR imaging ghost atomization in gas insulated switchgear,unclear texture,being easy to cause misjudgment and so on,this paper proposes an improved X-DR image detection algorithm for gas insulated switchgear(GIS)internal defects based on super-resolution attention mechanism.Based on the efficient sub-pixel convolutional neural network,two SE modules of computer attention mechanism are introduced to construct a novel SE-ESPCN super-resolution network.By evaluating the importance of X-DR image channels of GIS equipment,different proportions of image convolution layers are given to enhance the imaging effect of image fault details.The experimental results show that the fusion of SE module and sub-pixel convolutional neural network not only ensures the real-time performance of equipment fault area,but also makes the defect details of high-resolution X-DR image output by the algorithm more clear and visible,which is easy to observe,and improves the efficiency and accuracy of defect detection of staff.It has a certain reference value for the improvement of X-DR imaging system of GIS equipment in practical engineering.