Scholarlink

Remote sensing imagery is a primary source for land cover monitoring and assessment, for example, through infrastructure monitoring, urban planning, and disaster management, such as flooding and population growth. Deep learning techniques have recently gained significant attention due to their significant capabilities in LULC classification. This study proposes a CNN model to investigate the effect of patch size on LULC classification accuracy. The model was trained on multi-spectral data captured by the Sentinel-2 satellite for the Kut region of Iraq. This data was generated for patch sizes of 5 × 5, 7 × 7, and 64 × 64 using a QGIS plugin. The results indicate that the patch size 5 × 5 outperforms an accuracy of 97.6%, a kappa coefficient of 0.95, and an F1 score of 97% compared to patches 7 × 7 and 64 × 64.

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

The Effect of Patch Size on the Classification Accuracy of Multi-Spectral Remote Sensing Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Zainab Nasser Hussein,
  • Ehsan Ali Al-Zubaidi

摘要

Remote sensing imagery is a primary source for land cover monitoring and assessment, for example, through infrastructure monitoring, urban planning, and disaster management, such as flooding and population growth. Deep learning techniques have recently gained significant attention due to their significant capabilities in LULC classification. This study proposes a CNN model to investigate the effect of patch size on LULC classification accuracy. The model was trained on multi-spectral data captured by the Sentinel-2 satellite for the Kut region of Iraq. This data was generated for patch sizes of 5 × 5, 7 × 7, and 64 × 64 using a QGIS plugin. The results indicate that the patch size 5 × 5 outperforms an accuracy of 97.6%, a kappa coefficient of 0.95, and an F1 score of 97% compared to patches 7 × 7 and 64 × 64.