Scholarlink

Ground clutter contamination can be challenging for accurate wind estimation in wind profiler radars. Deep learning techniques can unlock nuanced wind behavior insights from complex spectral data, fostering advancements in atmospheric research. This paper primarily investigates the performance of three prominent deep learning architectures: GoogLeNet, SqueezeNet, and ResNet18, in classifying the wind profiler Doppler power spectrum contaminated by ground clutter. In the experimental evaluation, SqueezeNet got an accuracy of 99.7%, slightly outperforming the GoogLeNet and ResNet18 models that achieved an accuracy of 99.4% and 99.3%, respectively. The fivefold cross-validation results demonstrate the supremacy of the deep learning models. Furthermore, the explainability analysis unveiled the robustness and efficacy of the deep learning models in identifying ground clutter patterns within the Doppler power spectrum.

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

Exploring Deep Learning Models for Classifying Wind Profiler Doppler Power Spectrum Contaminated by Ground Clutter

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Baazil P. Thampy,
  • M. V. Judy,
  • Ajil Kottayil

摘要

Ground clutter contamination can be challenging for accurate wind estimation in wind profiler radars. Deep learning techniques can unlock nuanced wind behavior insights from complex spectral data, fostering advancements in atmospheric research. This paper primarily investigates the performance of three prominent deep learning architectures: GoogLeNet, SqueezeNet, and ResNet18, in classifying the wind profiler Doppler power spectrum contaminated by ground clutter. In the experimental evaluation, SqueezeNet got an accuracy of 99.7%, slightly outperforming the GoogLeNet and ResNet18 models that achieved an accuracy of 99.4% and 99.3%, respectively. The fivefold cross-validation results demonstrate the supremacy of the deep learning models. Furthermore, the explainability analysis unveiled the robustness and efficacy of the deep learning models in identifying ground clutter patterns within the Doppler power spectrum.