SPSNet: semantic-guided perspective shift network for robust person re-identification in drone imagery

Person re-identification using drone technology is increasingly important but faces challenges due to morphological compression and perspective distortions. This study introduces SPSNet, a novel framework that employs semantic-guided morphospatial encoding and decoding to mitigate these issues. SPSNet integrates a body spatial spotlight module to emphasize human features and a perspective shift shielding module to derive stable feature representations. Comprehensive experiments on four drone-based person ReID datasets demonstrate that SPSNet significantly outperforms state-of-the-art methods, achieving improvements of up to 11.5% in mAP and 11.0% in rank-1 accuracy on the PRAI dataset. Our approach facilitates the extraction of more distinct and stable features, making it well-suited for drone-based person re-identification tasks. Code is available at https://github.com/weihongwei3/SPSNet.