Scholarlink

<p>Timely and accurate information on socioeconomic development dynamics is crucial for effective government decision-making. However, traditional methods for acquiring socioeconomic indices often rely on time-consuming and labor-intensive economic censuses, limiting the quality and timeliness of the data. Recently, satellite remote sensing data, with its high temporal resolution and extensive coverage, has emerged as a promising tool for deriving economic development indices efficiently and precisely. In this paper, we propose a weakly supervised training framework that utilizes a large volume of unlabeled satellite imagery to estimate regional socioeconomic indices, thereby reducing reliance on labeled data and improving model reliability. Our framework outperforms traditional supervised methods, offering enhanced accuracy and interpretability. This provides novel insights into the factors driving socioeconomic dynamics, using satellite imagery as big data. We successfully apply the proposed framework to estimate a social development index across China and aggregate it into a county-level development index. Additionally, we adjust the county-level index to account for variations in the administrative region’s area. Our approach achieves superior performance, as demonstrated by Pearson correlation coefficients exceeding 0.8 with multiple economic indices using the same dataset. Furthermore, our interpretability analysis reveals that the model is highly sensitive to features such as roads and buildings, while showing limited sensitivity to farmlands. All codes and implementation details are publicly available at <a href="https://github.com/xiafire/Economic-Development">https://github.com/xiafire/Economic-Development</a>.</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

Weakly supervised deep learning for fine-grained socioeconomic development index inference based on satellite imagery

  • 期刊论文

  • 发表日期：

    2025年5月31日

查看全文
  • Yanbing Bai,
  • Bolin Zhang,
  • Ziyue Zhang,
  • Shubo Zeng,
  • Zhe Liu,
  • Xuxin Mao

摘要

Timely and accurate information on socioeconomic development dynamics is crucial for effective government decision-making. However, traditional methods for acquiring socioeconomic indices often rely on time-consuming and labor-intensive economic censuses, limiting the quality and timeliness of the data. Recently, satellite remote sensing data, with its high temporal resolution and extensive coverage, has emerged as a promising tool for deriving economic development indices efficiently and precisely. In this paper, we propose a weakly supervised training framework that utilizes a large volume of unlabeled satellite imagery to estimate regional socioeconomic indices, thereby reducing reliance on labeled data and improving model reliability. Our framework outperforms traditional supervised methods, offering enhanced accuracy and interpretability. This provides novel insights into the factors driving socioeconomic dynamics, using satellite imagery as big data. We successfully apply the proposed framework to estimate a social development index across China and aggregate it into a county-level development index. Additionally, we adjust the county-level index to account for variations in the administrative region’s area. Our approach achieves superior performance, as demonstrated by Pearson correlation coefficients exceeding 0.8 with multiple economic indices using the same dataset. Furthermore, our interpretability analysis reveals that the model is highly sensitive to features such as roads and buildings, while showing limited sensitivity to farmlands. All codes and implementation details are publicly available at https://github.com/xiafire/Economic-Development.