|
| 基于自注意力编码器和CNN的机动车鸣笛声识别* |
| Recognition of vehicle horn sounds based on self-attention encoder and CNN |
| 投稿时间：2024-10-21 修订日期：2024-12-18 |
| 中文摘要: |
| 为解决城市道路中违法鸣笛事件的识别和定位问题，本文提出了一种融合双输入自注意力编码器和卷积神经网络的机动车鸣笛识别方法。该方法通过结合自注意力机制的全局位置捕捉能力与卷积神经网络的局部特征挖掘能力，构建出具有高度判别性的声音特征。实验结果显示，所提方法在两个鸣笛数据集上的平均识别准确率分别达到90.2%和93.76%，在准确率方面明显优于现有鸣笛识别技术。此外，本文深入分析不同车辆尺寸类型的鸣笛喇叭声学特性，归纳了3种车辆尺寸类别，并在鸣笛声分类实验中取得了86.7%的平均准确率，验证了基于鸣笛声推断机动车尺寸的可行性。 |
| 英文摘要: |
| This paper proposes a vehicle honking recognition method that combines a dual-input self-attention encoder and a convolutional neural network to identify and locate illegal honking events. By leveraging the global position capturing ability of self-attention and the local feature extraction capacity of convolutional networks, the method extracts highly discriminative sound features. Results show that the proposed method achieves average accuracies of 90.2% and 93.76% on two datasets, outperforming existing whistle recognition methods. Additionally, the analysis of honking horn characteristics across three vehicle size categories achieves an 86.7% classification accuracy, confirming the feasibility of inferring vehicle size from honking sounds. |
| DOI： |
| 中文关键词: 机动车鸣笛声识别 双输入自注意力编码器 卷积神经网络 特征融合 车辆尺寸 |
| 英文关键词: Identification of vehicle horn sound Self attention coder Convolutional neural network Feature extraction vehicle dimension |
| 基金项目: |
|
| 摘要点击次数: 11 |
| 全文下载次数: 0 |
|
查看/发表评论 下载PDF阅读器 |
| 关闭 |
