Scholarlink

Forest fires, a significant global threat causing vast annual devastation where each year millions of hectares are destroyed over the world. Burned areas necessitate early and accurate detection for effective response. This study investigates the application of Feature Selection (FS) to address data complexity in forest fire detection utilizing PRISMA hyperspectral imagery (HSI). FS is employed to select the most relevant information, thereby reducing complexity and facilitating the interpretation of results. The effectiveness of Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), Random Forest (RF), and Support Vector Machine (SVM) algorithms is investigated. Analysis reveals that while the raw 229-band data yielded a reasonable accuracy with RF, applying FS through wrapper methods only select 13 bands and improve the performance of LGBM, XGBoost, and SVM. Notably, the best model achieves a classification precision of 94% and recall of 91%, demonstrating the effectiveness of FS in enhancing forest fire detection and mapping.

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

Enhancing Forest Fire Classification with Feature Selection and Machine Learning Based on PRISMA Hyperspectral Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Hanifi Meroufel,
  • Faouzi Berrichi,
  • Mohammed Bilel Amri,
  • Mohammed El Amin Larabi

摘要

Forest fires, a significant global threat causing vast annual devastation where each year millions of hectares are destroyed over the world. Burned areas necessitate early and accurate detection for effective response. This study investigates the application of Feature Selection (FS) to address data complexity in forest fire detection utilizing PRISMA hyperspectral imagery (HSI). FS is employed to select the most relevant information, thereby reducing complexity and facilitating the interpretation of results. The effectiveness of Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), Random Forest (RF), and Support Vector Machine (SVM) algorithms is investigated. Analysis reveals that while the raw 229-band data yielded a reasonable accuracy with RF, applying FS through wrapper methods only select 13 bands and improve the performance of LGBM, XGBoost, and SVM. Notably, the best model achieves a classification precision of 94% and recall of 91%, demonstrating the effectiveness of FS in enhancing forest fire detection and mapping.