arXiv cs.AI· Junseo Min, Joongheon Kim·· 12 小时前AI 评分18
StructAttn：面向细粒度毫米波液体识别的结构化 Token 注意力
Structured-Token Attention for Fine-Grained mmWave Liquid Identification
AI 导读
StructAttn 用结构化 Token 注意力将雷达特征张量表示为天线、特征、分支与全局四个互补视图，嵌入 11 个 token 后接入三层 Transformer，在公开 FG-LiquID 数据集（30 种液体、3000 个样本）上取得 96.70% 的平均最佳 epoch 准确率，比标量 token 注意力模型高 1.50 个百分点。
正文 · 原文
Abstract:Fine-grained liquid identification requires a classifier to distinguish similar substances from compact millimeter-wave measurements. We investigate structured-token attention, which represents a radar feature tensor through complementary antenna, feature, branch, and global views. The proposed StructAttn model embeds these views into eleven tokens and combines them with a three-layer transformer. Experiments on the public FG-LiquID data cover 30 liquids and 3,000 samples. Over five seeds on the released split, StructAttn achieves 96.70\% mean best-epoch accuracy, matching the original-recipe MMNet baseline and exceeding a scalar-token attention model by 1.50 percentage points. A five-seed probability ensemble reaches 97\%, corresponding to 18 errors among 600 test samples. Final-epoch and five-fold results complement the checkpoint-selected comparison, with StructAttn obtaining 95.30\% out-of-fold accuracy. Token and feature analyses show that structured views are useful for prediction, while residual confusions concentrate in water, soft-drink, and edible-oil classes. These results support structured tokenization as a practical design choice for attention-based liquid recognition.
| Comments: | ICEIC 2027 paper preprint |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.27512 [cs.AI] |
| (or arXiv:2610.27512v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.27512 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Junseo Min [view email]
[v1]
Tue, 20 Oct 2026 02:27:39 UTC (84 KB)
来源：arXiv cs.AI · arxiv.org