Aiming at the problems of poor recognition accuracy and low recognition rate of group actions in volleyball sports， an intelligent volleyball action recognition technology is proposed. The technology uses multi fully convolutional networks （MFCN） to extract the feature data of the antheletes and corrects the model through the Markov random field. Then， the dual attention model is introduced to strengthen the focus on feature attention and reduce the problem of key feature data loss. Using the model to test volleyball action recognition， the results show that the accuracy rate of single-person volleyball action recognition is 0.986， and the recognition accuracy and convergence speed are high. The accuracy rate of group volleyball action recognition is 0.962， which is better than that of similar models. The research model has excellent results in practical scene applications， and can provide a technical reference for the recognition of sports actions and the improvement of visual technologies..