28 January 2025 in IEEE Access
Pedestrian re-identification is influenced by various complex factors such as camera angles, posture changes, and object occlusion. How to effectively extract discriminative feature information remains a key challenge. To address this issue, this paper proposes an enhanced neural network based on residual attention mechanism and random erasure. Firstly, by introducing attention mechanism in the residual network, the response of key features is effectively strengthened and redundant features are suppressed, thereby improving the discriminative performance of the network. Secondly, using random erasure as a data augmentation method effectively alleviates the impact of occlusion on the model’s recognition ability, reduces the risk of overfitting, and improves the network’s generalization performance. In addition, using triplet loss for supervised training of the network promotes better clustering performance of samples in the feature space, thereby significantly improving recognition accuracy. The experimental results show that the proposed method achieves a Rank-1 accuracy of 88.55% and an average accuracy (mAP) of 79.45% on the Market-1501 dataset, which is superior to existing methods. Verified that the proposed method can better extract discriminative pedestrian features and improve the performance of pedestrian re-identification models.