The visual location system based on marker are mainly designed for land. There are few marker identification algorithms designed for the underwater environment, and there has been no identification algorithms for the under-ice environment. This paper proposes an identification algorithm of visual location Mark under ice based on parallel computing, which includes two parts: line detection and line combination. This algorithm has a more thoughtful line combination strategy for better adaption to under-ice environment. And it solve the problem of longer run time caused by complex strategy through parallel computing optimization. Due to the operations of identification can be divide into multiply sub-task, the algorithm can be accelerated by GPU. Besides, the image processing method of the algorithm is more efficient than deep neural network, which need lower computing power. The test of the algorithm in pools and under ice shows that the effective distance and speed are better than the existing mature algorithms.