Person re-identification aims to retrieve specific person objects across surveillance cameras. Due to problems such as attitude changes, object occlusion, and background interference, the person re-identification effect is not good. An unsupervised video person re-identification method using multi-core expanded convolution is proposed to make full use of the extracted person features. First, the pre-trained ResNet50 is used as the encoder, and the multi-core expanded convolution module is introduced into the encoder to increase the convolution kernel receptive field to capture local and global features, and obtain more semantic information through the decoder to enhance feature representation; Secondly, a multi-scale feature fusion module is introduced at the output of the decoder to fuse the features in adjacent layers, further reducing the semantic gap between different feature channel layers to produce more robust feature representations; finally, in PRID2011, iLIDS-VID Experimental verification with mainstream datasets such as DukeMTMC-VideoReID shows that the improved model has higher performance indicators and recognition accuracy, and can better express person characteristics.