华中科技大学学报（自然科学版）2025，Vol.53Issue(5)：128-134,142,8.DOI:10.13245/j.hust.250539
基于足迹高频特征的多任务人体生物特征估计
Multi-task human biometric estimation based on high-frequency features of footprints
摘要
Abstract
To address the task conflict issues in existing multi-task joint training methods for estimating multiple human biometrics from footprint images,a high-frequency feature multi-task network(HiFM-Net)for multi-task human biometric estimation based on footprint high-frequency features was proposed.First,considering the characteristics of small inter-class differences and large intra-class differences in footprint images,a high-frequency feature branch based on multi-scale Laplacian convolution kernels was designed to extract high-frequency features such as footprint contours and heavy pressure areas.Secondly,to solve the multi-task conflict problem,a multi-task interaction branch combining a soft-sharing structure and an attention mechanism was developed to extract task-specific features while mining and utilizing the correlations between tasks.Then,a fusion module was constructed to orthogonally integrate both footprint shared features and high-frequency features with task-specific features respectively,thereby enhancing the detail representation capability of each task feature.Finally,experiment results conducted on the optical barefoot footprint dataset show that the multi-task comprehensive evaluation metric is improved by 6.48%,while the C3 accuracy rates for height and weight estimation tasks are increased by 6.25%and 8.26%respectively compared with single-task models,indicating that the proposed method effectively mitigates multi-task conflicts in footprint analysis and enhances individual task performance.关键词
足迹图像/生物特征估计/多任务交互/高频信息/特征融合Key words
footprint image/biometric estimation/multi task interaction/high frequency information/feature fusion分类
信息技术与安全科学引用本文复制引用
张艳,粟娇娇,王科,王年..基于足迹高频特征的多任务人体生物特征估计[J].华中科技大学学报（自然科学版）,2025,53(5):128-134,142,8.基金项目
安徽省重点研究与开发计划科技合作专项科技强警项目(2022k07020006) （2022k07020006）
安徽高校自然科学研究重大项目(KJ2021ZD0004) （KJ2021ZD0004）
安徽高校协同创新资助项目(GXXT-2022-038). （GXXT-2022-038）
