Use of Unmanned Aerial Vehicle Imagery and Deep Learning UNet to Extract Rice Lodging

Area of Science:

• Agricultural Science
• Remote Sensing
• Computer Vision

Background:

• Rice lodging significantly reduces harvest yields, and traditional assessment methods are inefficient.
• Manual on-site measurements for rice lodging are time-consuming, labor-intensive, and costly.

Purpose of the Study:

• To develop and evaluate a novel, efficient method for rice lodging assessment using deep learning.
• To compare the effectiveness of Red, Green, and Blue (RGB) and multispectral imagery for rice lodging detection.

Main Methods:

• Utilized an Unmanned Aerial Vehicle (UAV) with RGB and multispectral cameras to capture images of lodged and non-lodged rice fields.
• Developed a deep learning UNet (U-shaped Network) model trained on augmented image datasets.
• Spliced and cropped original images to create specific datasets for model training and validation.

Main Results:

• The UNet model achieved high accuracy, with Dice coefficients of 0.9442 for RGB images and 0.9284 for multispectral images.
• RGB images, without feature extraction, demonstrated superior performance in recognizing rice lodging compared to multispectral images.

Conclusions:

• The proposed deep learning approach offers an efficient and cost-effective solution for large-area rice lodging monitoring.
• This method provides a valuable tool for agricultural research and practical crop management, improving upon traditional assessment techniques.