Due to the complex underwater environment, underwater images are usually degraded low-quality images. Therefore, a multi attention mechanism guided binocular image super-resolution reconstruction algorithm is proposed to selectively learn image feature information for achieving high-quality image reconstruction. Aiming at the low resolution of underwater image, a network with double attention module is designed to enhance the learning of important details; Then, aiming at the disparity characteristics of binocular images, a parallax attention module is proposed to fully learn the prior information of left and right-hand images, and improve the image quality effectively. The PSNR of the reconstructed image with x2 and x4 on the Middlebury dataset is 33.3dB and 28.39dB respectively. It shows that the algorithm can improve the spatial resolution of the image and better retain the image details. At the same time, the reconstruction effect of this algorithm is better than other algorithms on the underwater dataset in real underwater scenes, indicating that it can achieve higher quality underwater image super-resolution reconstruction.