<![CDATA[Person recognition under varied conditions is a critical task that aims to accurately identify individuals captured from different angles or at different times. One of the primary challenges in this field is occlusion, which significantly degrades recognition performance. To address this issue, we propose an advanced attention-based network designed to mitigate the effects of occlusion and enhance recognition accuracy. Our approach leverages channel attention to dynamically recalibrate the importance of each channel, utilizing both width and depth attention mechanisms to emphasize discriminative and informative features. The network employs a multi-scale feature extraction strategy, partitioning feature maps to capture multi-level representations of the human body. The concatenation of results from these attention stages facilitates the integration of local and global features, effectively reducing the impact of occlusion. We evaluate the proposed model on multiple benchmark datasets, including PRID 2011, iLIDS-VID, and Market-1501. The experimental results demonstrate that our model achieves superior performance, attaining a top accuracy of 99.79% on the PRID 2011 dataset, 98.55% on the iLIDS-VID dataset, and 88.24% on the Market-1501 dataset.]]>
Copyrights © 2026
