Flower skating action detection algorithm based on improved 3DSE attention mechanism

The invention belongs to the technical field of image processing and computer vision, and particularly relates to a figure skating action detection algorithm based on an improved 3DSE attention mechanism, the method is based on a YOWO network, and the network inputs RGB video continuous frame and RGB key frame information to a time sequence detection module and a space detection module for processing; in a time sequence detection module, an attention mechanism 3DSE module is introduced to improve the sensitivity of a model to channel features, and continuous 3D depth separable convolution and residual connection are added in front of a pooling layer of a 3DSE attention mechanism, so that the network pays more attention to feature information of a shallow layer and a deep layer, it is ensured that key information is reserved, and the detection accuracy is improved. Therefore, the spatial-temporal feature extraction capability of the 3D convolutional network is improved; in the space detection module, carrying