Target detection, recognition, and tracking technology has been widely concerned with the development of artificial intelligence. However, images or videos taken from a long distance could be easily affected by atmospheric turbulence so degradation often occur. With the aim of providing reliable data for further processing, a new method is proposed to obtain sharp image and video from those with aero-optical effect. Firstly, both the degradation and reconstruction models are established. Secondly, mean filter is applied to remove the additive noise in the shooting process of infrared camera. Then, combined with the pyramid structure, the generated countermeasure network based on MobileNet-v2 is used to restore the blurred image without noise. Finally, experiments with the evaluation indexes including Peak-Signal to Noise Ratio (PSNR), Structural Similarity (SSIM), and Kernelized Correlation Filters (KCF) are carried out to demonstrate the significant improvements over competitive algorithms.