Vehicle Re-Identification Method Based on Efficient Self-Attention CNN-Transformer and Multi-Task Learning Optimization

To address the challenges of low accuracy in vehicle re-identification caused by intra-class variations, inter-class similarities and environmental factors, this paper proposes a CNN-Transformer architecture (IBNT-Net) for vehicle re-identification. The method builds upon a ResNet50-IBN backbone network and incorporates an improved multi-head self-attention mechanism to aggregate contextual information. It constructs a multi-branch vehicle re-identification network that combines both global and local features. Furthermore, a multi-task learning strategy is adopted, creating specialized learning pathways for classification tasks and metric learning tasks. Group convolution techniques are utilized to reduce model complexity, making it suitable for resource-constrained environments. On the VeRi-776 and VehicleID dataset, the proposed method achieves state-of-the-art performance with less parameters. The experimental results show that the proposed method has better re-identification performance and the extracted features are more discriminative.