Scholarlink

<p>To address the problem of poor anti-smoke interference in traditional laser ranging algorithms, this paper proposes a novel distance detection method that integrates Time Series Concatenation, Convolutional Neural Networks, and Customized Non-Maximum Suppression based on deep learning. First, the TSC technique is applied to convert the one-dimensional echo waveform time series into a two-dimensional representation, preserving both the temporal dependencies and the potential relationships between data points. Next, a Convolutional Neural Networks is employed to extract key features—such as category, position, strength, confidence, and pulse width—from the TSC-concatenated image. Finally, Customized Non-Maximum Suppression is used to accurately identify the target and estimate its distance. By selecting the appropriate pulse width, sampling rate, and convolutional layer configurations, the proposed method achieves an F1 recognition score of 93.78%, a false alarm rate of 6.12%, and a ranging accuracy of 0.055&#xa0;m. The results demonstrate that the TSC-CNN-CNMS approach outperforms traditional algorithms in terms of anti-smoke interference resistance and ranging accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detection of pulsed laser in smoke based on time series concatenation and convolutional neural networks

  • 期刊论文

  • 发表日期：

    2025年11月3日

查看全文
  • Xing Ding,
  • Bo Liu,
  • Chao Liang,
  • Lanjie Niu

摘要

To address the problem of poor anti-smoke interference in traditional laser ranging algorithms, this paper proposes a novel distance detection method that integrates Time Series Concatenation, Convolutional Neural Networks, and Customized Non-Maximum Suppression based on deep learning. First, the TSC technique is applied to convert the one-dimensional echo waveform time series into a two-dimensional representation, preserving both the temporal dependencies and the potential relationships between data points. Next, a Convolutional Neural Networks is employed to extract key features—such as category, position, strength, confidence, and pulse width—from the TSC-concatenated image. Finally, Customized Non-Maximum Suppression is used to accurately identify the target and estimate its distance. By selecting the appropriate pulse width, sampling rate, and convolutional layer configurations, the proposed method achieves an F1 recognition score of 93.78%, a false alarm rate of 6.12%, and a ranging accuracy of 0.055 m. The results demonstrate that the TSC-CNN-CNMS approach outperforms traditional algorithms in terms of anti-smoke interference resistance and ranging accuracy.