ALPD-Net: a wild licorice detection network based on UAV imagery.

Area of Science:

• Botany
• Ecology
• Remote Sensing
• Computer Vision

Background:

• Wild licorice is ecologically and medicinally important but threatened by overharvesting.
• Traditional survey methods are inadequate for large-scale monitoring in complex terrains.
• Accurate distribution and growth condition data are crucial for wild licorice conservation.

Purpose of the Study:

• To develop a novel method for precise wild licorice identification and distribution mapping.
• To address the limitations of traditional survey methods using Unmanned Aerial Vehicle (UAV) technology.
• To create a robust deep learning model for detecting wild licorice in diverse environments.

Main Methods:

• A new dataset of wild licorice images was collected using UAVs.
• A novel detection network, ALPD-Net, was proposed, incorporating an Adaptive Background Suppression Module (ABSM) and a Lightweight Multi-Scale Module (LMSM).
• A Progressive Feature Fusion Module (PFFM) was developed to enhance feature fusion and reduce information loss.

Main Results:

• ALPD-Net achieved high detection accuracy with 73.3% precision, 76.1% recall, and 79.5% mAP50.
• The model demonstrated superior performance compared to mainstream object detection models in identifying wild licorice.
• ALPD-Net significantly reduced missed and false detections, improving reliability for surveys.

Conclusions:

• ALPD-Net offers a promising solution for accurate and efficient large-scale wild licorice monitoring using UAV remote sensing.
• The developed network effectively handles complex backgrounds and varying licorice scales.
• This technology supports sustainable management and conservation of valuable wild licorice resources.