Vehicle Reidentification Based on Multi-Attribute Adaptive Polymerization Network

To address the insufficient feature perception ability and reduced recognition accuracy in vehicle reidentification due to intra-class differences and inter-class similarity of vehicle targets, a vehicle reidentification method based on multi-attribute adaptive aggregation network (MaAPN) architecture is proposed. First, the ResNet-50 network is used as the backbone network for feature extraction, and the instance-batch normalization (IBN) adaptive module is introduced to extract feature representations with strong domain adaptability. Next, the attributes such as camera perspective, vehicle type, and vehicle color are integrated into the network to develop a multi-attribute self-attention feature enhancement model to enhance the robustness and discriminability of feature representation. Finally, a comprehensive loss function is designed to further improve the accuracy of the network by optimizing the feature distance between samples. The results show that the MaAPN architecture achieves an average accuracy of 87.3% and 86.9% on the VeRi-776 and VERI-Wild datasets respectively, and achieves optimal results on various evaluation indicators, effectively improving the accuracy of vehicle reidentification.