Scholarlink

This study illustrates the efficacy of the Self-Organizing Map (SOM) algorithm within the field of seismology, specifically focusing on waveform quality assessment and mislabelled catalog information. Employing Seismic data from the Campi Flegrei caldera, this study applies Linear Predictive Coding (LPC) to preprocess the waveforms before SOM training. This method successfully reveals distinct waveform clusters, indicating alterations in the seismic station during a specific period. A second application scrutinizes over 160, 000 first-motion polarities (FMP) from the Italian seismic dataset for machine learning (INSTANCE), employing Principal Component Analysis (PCA) alongside SOM to identify and exclude erroneous labels. This resulted in a cleaner and more reliable dataset by removing approximately 10, 000 waveforms with no reliable FMP label. These approaches underscore the potential of SOM in improving dataset quality by identifying artifacts and mislabelled examples, thereby optimizing seismic data analysis.

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

Applications of SOM in Seismology: from Dataset Quality Assessment to Mislabelled Trace Detection

  • 章节

  • 发表日期：

    2026年

查看全文
  • Alberico Grimaldi,
  • Giovanni Messuti,
  • Ortensia Amoroso,
  • Ferdinando Napolitano,
  • Paolo Capuano,
  • Vincenzo Convertito,
  • Danilo Galluzzo,
  • Silvia Scarpetta

摘要

This study illustrates the efficacy of the Self-Organizing Map (SOM) algorithm within the field of seismology, specifically focusing on waveform quality assessment and mislabelled catalog information. Employing Seismic data from the Campi Flegrei caldera, this study applies Linear Predictive Coding (LPC) to preprocess the waveforms before SOM training. This method successfully reveals distinct waveform clusters, indicating alterations in the seismic station during a specific period. A second application scrutinizes over 160, 000 first-motion polarities (FMP) from the Italian seismic dataset for machine learning (INSTANCE), employing Principal Component Analysis (PCA) alongside SOM to identify and exclude erroneous labels. This resulted in a cleaner and more reliable dataset by removing approximately 10, 000 waveforms with no reliable FMP label. These approaches underscore the potential of SOM in improving dataset quality by identifying artifacts and mislabelled examples, thereby optimizing seismic data analysis.