[论文] HumP-KD: A Hybrid Uncertainty-Aware Multi-Stage Progressive Knowledge ...

论文概要

研究领域: CV 作者: Mohammed Arif Mainuddin, Najifa Tabassum, Omar Ibne Shahid 发布时间: 2026-06-12 arXiv: 2606.14684

中文摘要

实时火灾分类系统需要同时满足准确、计算高效且可在资源受限硬件上部署的模型。本工作提出了HumP-KD，一种混合不确定性感知多阶段渐进知识蒸馏框架，用于高效火灾分类。使用两个数据集，FlameVision和Dataset-II，分别包含8,600和31,309张图像。在标准预处理、在线增强、高斯噪声和运动模糊鲁棒性条件下应用各种CNN和变换器基线。所提出的HumP-KD模型通过三个紧密集成组件，将来自两个冻结异构变换器教师Swin-Tiny和ViT-Base及其Meta-MLP集成体的知识蒸馏到轻量级MobileViT-S学生模型。层次渐进知识蒸馏使用层次特征构建器，生成融合的空间注意力掩模以选择性地引导蒸馏到判别区域。多阶段知识蒸馏在训练过程中渐进激活三个蒸馏阶段。在Dataset-II上，HumP-KD在10次独立试验中达到平均F1分数0.9876 ± 0.0063，显著优于无蒸馏训练的MobileViT-S基线（0.9537 ± 0.0351），统计显著性由独立t检验（p = 0.0195）和Wilcoxon符号秩检验（W = 1, p = 0.0039）确认。所提出的方法还展示了跨数据集的强泛化能力和在退化视觉条件下的鲁棒性。学生模型仅保留4.94M参数和19.01Mb模型大小，比Swin-Tiny减少5.7倍参数，比ViT-Base减少17.5倍，同时实现37.72 CPU FPS，适合实时部署。

原文摘要

Real-time fire classification systems require models that are simultaneously accurate, computationally efficient, and deployable on resource-constrained hardware. This work proposes HumP-KD, a Hybrid Uncertainty-aware Multi-stage Progressive Knowledge Distillation framework for efficient fire classification. Two datasets, FlameVision and Dataset-II, containing 8,600 and 31,309 images, are used. Various CNN and transformer baselines are applied under standard preprocessing, online augmentation, Gaussian noise and motion blur robustness conditions. The proposed HumP-KD model distills knowledge from two frozen heterogeneous transformer teachers, Swin-Tiny and ViT-Base, along with their Meta-MLP ensemble, into a lightweight MobileViT-S student via three tightly integrated components. Hierarchi...

--- *自动采集于 2026-06-16*

#论文 #arXiv #CV #小凯