Scholarlink

<p>Monitoring human activities in remote areas presents significant challenges due to lacking communication networks and infrastructure. In this context, using publicly available satellite imagery offers a cost-effective solution, as it enables the identification of changes in these areas. However, specific scenarios make detection more complicated. One such scenario is detecting indoor activity within buildings in remote areas. Walls and roofs create barriers for most sensors. Nevertheless, activities inside buildings can be associated with heat emissions, which specific remote sensors can detect. Unfortunately, publicly available satellite data does not include information from such sensors. In light of this limitation, this study investigates the opportunity of using machine learning models to interpret public-available data. Specifically, we trained four machine learning models (XGBoost, LGBM, DNN, and CNN) using images from Sentinel-2 Band 12 (the sensor with the frequency range closest to the heat emission peak) and meteorological data (temperature). Our results show that these models can identify farm-building activity, with the XGBoost model achieving the highest accuracy of 0.96 by integrating satellite data and temperature information; the findings suggest that leveraging public satellite sensors can effectively detect human heat emissions and improve surveillance in remote areas, overcoming some limitations of traditional methods.</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 changes in the heat emissions signature of buildings related to indoor activity using publicly available satellite data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月30日

查看全文
  • Mario E. Suaza-Medina,
  • Javier Lacasta,
  • Francisco J. López-Pellicer,
  • Rubén Béjar,
  • F. Javier Zarazaga-Soria

摘要

Monitoring human activities in remote areas presents significant challenges due to lacking communication networks and infrastructure. In this context, using publicly available satellite imagery offers a cost-effective solution, as it enables the identification of changes in these areas. However, specific scenarios make detection more complicated. One such scenario is detecting indoor activity within buildings in remote areas. Walls and roofs create barriers for most sensors. Nevertheless, activities inside buildings can be associated with heat emissions, which specific remote sensors can detect. Unfortunately, publicly available satellite data does not include information from such sensors. In light of this limitation, this study investigates the opportunity of using machine learning models to interpret public-available data. Specifically, we trained four machine learning models (XGBoost, LGBM, DNN, and CNN) using images from Sentinel-2 Band 12 (the sensor with the frequency range closest to the heat emission peak) and meteorological data (temperature). Our results show that these models can identify farm-building activity, with the XGBoost model achieving the highest accuracy of 0.96 by integrating satellite data and temperature information; the findings suggest that leveraging public satellite sensors can effectively detect human heat emissions and improve surveillance in remote areas, overcoming some limitations of traditional methods.