An intelligent recognition method for irregular movements of sports players based on improved fish swarm algorithm

This paper proposes an intelligent recognition method based on an improved fish swarm algorithm to address complex issues such as noise, occlusion, and viewpoint changes in sports athlete movement data. The aim is to achieve stable and accurate recognition of non-standard movements in complex scenes. Firstly, the preprocessing module is used to denoise multi view action videos. Subsequently, the 3D ResNet module was used to extract spatiotemporal local features of non-standard actions, and their temporal features were further extracted through multi kernel convolution LSTM. The core innovation of this article lies in the introduction of an improved fish swarm algorithm, which combines a clustering based initialization strategy with an escape mechanism that introduces crossover and mutation operations, effectively optimizing the parameters of the multi kernel convolutional LSTM and significantly enhancing the robustness and recognition accuracy of the model in complex situations such as noise, occlusion, and viewpoint changes. Finally, combined with the SoftMax classifier, the recognition results of non-standard actions are outputted. The experimental results show that when the target occlusion rate is 5% and 30%, the maximum recognition errors of this method are only 2% and 4%, respectively; In tests covering 5 types of sports, the highest recognition probability was 0.91, successfully identifying various typical non-standard movements such as trunk misalignment, walking violations, and wrist bending.