Scholarlink

<p>Fusion of heterogeneous sensor signals offers a promising solution to early Red Palm Weevil (RPW) infestation detection in palm trees. In this work, we present a multimodal detection system that combines acoustic signals from larval activity with microwave backscatter signals to detect the presence of Red Palm Weevil inside a palm tree. The acoustic data is processed using a spectral feature extraction technique, while microwave signals are analyzed using a frequency domain feature extraction to detect shifts that indicate the presence of RPW. Each model undergoes independent feature extraction and classification using lightweight, efficient deep learning models, including an attention-guided fusion classifier and a sensor-guided transformer. To enhance decision reliability, a late fusion approach based on adaptive sequential Kalman filtering is used to fuse the outputs of the individual classifiers. Experimental tests on real-world datasets achieve 96.2% accuracy, demonstrating significant improvement in sensitivity and robustness compared with unimodal systems. The proposed system offers a non-invasive, field-deployable solution for precision pest monitoring and timely intervention in palm plants.</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

Multi-modal fusion of acoustic and microwave signals for early detection of Red Palm Weevil infestation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月23日

查看全文
  • Navya Stephen,
  • Betty Martin,
  • A. Rajesh

摘要

Fusion of heterogeneous sensor signals offers a promising solution to early Red Palm Weevil (RPW) infestation detection in palm trees. In this work, we present a multimodal detection system that combines acoustic signals from larval activity with microwave backscatter signals to detect the presence of Red Palm Weevil inside a palm tree. The acoustic data is processed using a spectral feature extraction technique, while microwave signals are analyzed using a frequency domain feature extraction to detect shifts that indicate the presence of RPW. Each model undergoes independent feature extraction and classification using lightweight, efficient deep learning models, including an attention-guided fusion classifier and a sensor-guided transformer. To enhance decision reliability, a late fusion approach based on adaptive sequential Kalman filtering is used to fuse the outputs of the individual classifiers. Experimental tests on real-world datasets achieve 96.2% accuracy, demonstrating significant improvement in sensitivity and robustness compared with unimodal systems. The proposed system offers a non-invasive, field-deployable solution for precision pest monitoring and timely intervention in palm plants.