Fuzzy Attribute Information Embedding to ViT for Aircraft Re-Identification1

Aircraft re-identification is a critical task in aviation and surveillance applications. This paper proposes a Fuzzy Attribute Information Embedding (FAIE) module to enhance aircraft re-identification by integrating textual attribute descriptions with visual data. The FAIE module employs a fuzzy self-attention mechanism to improve the model’s ability to focus on and distinguish attribute features. This method is integrated into the Transformer-based TransReID framework, utilizing multimodal data (image patches, auxiliary information, and attribute embeddings) to achieve feature representation. Experiments on Market-1501 and DukeMTMC-reID pedestrian datasets demonstrate state-of-the-art performance (98.50% mAP and 99.60% Rank-1 on Market-1501). For aircraft re-identification, with optimal attribute embedding strength equal to 0.6, the model achieves 99.2% mAP and 99.3% Rank-1. This method demonstrates the effectiveness of a comprehensive utilization of attribute text and image features in re-identification tasks.