Monitoring the leaf damage by the rice leafroller with deep learning and ultra-light UAV.

Area of Science:

• Agricultural Entomology
• Remote Sensing
• Computer Vision

Background:

• Rice leafroller infestation poses a significant threat to rice production, necessitating effective pest management strategies.
• Accurate monitoring of rice leafroller damage is hindered by limitations in image quality and identification methods.
• Existing studies on fast and accurate identification of rice leafroller damage are scarce.

Purpose of the Study:

• To develop a fast and accurate method for identifying and quantifying rice leafroller damage in rice fields.
• To leverage ultra-light unmanned aerial vehicles (UAVs) for high-resolution image acquisition of damaged rice areas.
• To apply deep learning segmentation models for precise damage area recognition and patch counting.

Main Methods:

• Utilized an ultra-light UAV to capture high-resolution images, mitigating downwash flow field interference.
• Employed the Attention U-Net deep learning segmentation model for recognizing rice leafroller-damaged areas.
• Developed a method to count damaged patches based on the segmented areas.

Main Results:

• Attention U-Net achieved a high F1 score of 0.908, outperforming the traditional Random Forest (RF) method.
• The UAV and deep learning approach demonstrated reasonable accuracy in identifying damage patches, with a coefficient of determination of 0.879.
• Ground validation confirmed the method's effectiveness, highlighting the uneven spatial distribution of damage.

Conclusions:

• Presents an efficient vision for monitoring rice leafroller damage using ultra-light UAVs.
• Contributes to effective control and management strategies for the hazardous rice leafroller pest.
• Highlights the potential of integrated UAV and deep learning technologies in agricultural pest monitoring.