ProFi-Net: Prototype-based Feature Attention with Curriculum Augmentation for WiFi-based Gesture Recognition
Zhe Cui, Shuxian Zhang, Kangzhi Lou, Le-Nam Tran
TL;DR
This work addresses the data-scarce regime of WiFi-based gesture recognition by introducing ProFi-Net, a few-shot learning framework that fuses prototype-based metric learning with a feature-level attention mechanism and a curriculum-inspired query augmentation strategy. The method uses a CNN to map CSI data into embeddings, computes class prototypes, and refines distances via an attention vector so $d(x_q, C_c)=A_c \cdot \| z_q - C_c \|^2$, paired with the standard cross-entropy objective. A curriculum-style augmentation on the query set progressively introduces Gaussian noise, quantified by $ ext{SNR}$, to improve generalization. Extensive experiments across three real-world environments show that ProFi-Net outperforms conventional prototype networks and other state-of-the-art few-shot methods in both accuracy and training efficiency, demonstrating strong robustness to environmental variations and feature sparsity in CSI data.
Abstract
This paper presents ProFi-Net, a novel few-shot learning framework for WiFi-based gesture recognition that overcomes the challenges of limited training data and sparse feature representations. ProFi-Net employs a prototype-based metric learning architecture enhanced with a feature-level attention mechanism, which dynamically refines the Euclidean distance by emphasizing the most discriminative feature dimensions. Additionally, our approach introduces a curriculum-inspired data augmentation strategy exclusively on the query set. By progressively incorporating Gaussian noise of increasing magnitude, the model is exposed to a broader range of challenging variations, thereby improving its generalization and robustness to overfitting. Extensive experiments conducted across diverse real-world environments demonstrate that ProFi-Net significantly outperforms conventional prototype networks and other state-of-the-art few-shot learning methods in terms of classification accuracy and training efficiency.