In light of the accuracy and timeliness of locomotion-mode recognition of intelligent prosthesis, a pattern recognition method based on the fish swarm (FA) algorithm is proposed to optimize the extreme learning machine (ELM). First, the features of tensor projection are extracted, and the rationality of the selection of features is analyzed. Then, principal component analysis is used to reduce the dimension. Finally, treads of walking on-ground, upstairs, downstairs, uphill, and downhill are recognized using the learning machine of FA-ELM. The recognition accuracy is 97.45%. The comparison with ELM and other classifiers shows that FA-ELM recognizes more accurately.