In recent years, deep learning has made a great achievement in image super-resolution reconstruction. Due to the complex ocean environment, the traditional image super-resolution algorithm has some problems, such as the difficulty in adjusting parameters. In addition, the single frame image super-resolution algorithm has an ill conditioned recovery and the generated pixels are uncertain. In this paper, a super-resolution reconstruction algorithm of multi-frame images is proposed for the study of sea surface image reconstruction. The convolution neural network in deep learning is used to learn the mapping relationship between multi-frame low-resolution images and high-resolution images, so as to realize super-resolution reconstruction. At the same time, because the ocean monitoring imaging system needs more high-frequency information to identify the target and lock contour, the residual network framework is proposed to improve the quality of network reconstruction images, recover more high-frequency information and enrich the image details. The experimental results show that the proposed algorithm has a better image reconstruction ability and better subjective and objective evaluation results compared with other methods.