A super-resolution reconstruction method based on local and global information attention generative adversarial network (LGIA-GAN) is proposed for low-resolution and ambiguous temperature distribution of infrared images of power equipment. Firstly

a detail-enhanced fusion convolution (DEFConv) is constructed by fusing multiple convolutional outputs using a gated weight unit to increase the proportion of important information in the output feature map. Secondly

a dual attention module (DAB) is built to model image long-range pixel dependencies and capture spatial and channel dimensional information. Thirdly

a generative adversarial network is constructed so that the network focuses on both local texture details and global contour information of the infrared image of the power equipment. Finally

it is experimentally demonstrated that the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of LGIA-GAN on the dataset are 30.266 dB and 0.919 7

respectively

and the reconstruction time is 0.120 s

which is significantly better than several other GAN algorithms

and the reconstruction effect is better in subjective vision. The proposed method can effectively improve the thermal imaging resolution of power equipment

which is supportive of power equipment fault diagnosis.