Rice Plant Counting, Locating, and Sizing Method Based on High-Throughput UAV RGB Images.

Area of Science:

• Agricultural Science
• Computer Vision
• Remote Sensing

Background:

• Manual rice plant counting is labor-intensive and time-consuming.
• Accurate plant counts are vital for yield estimation, growth monitoring, and disaster assessment in rice production.

Purpose of the Study:

• To develop an automated, efficient, and accurate method for rice plant counting, locating, and sizing.
• To reduce the workload associated with traditional manual counting methods.

Main Methods:

• Utilized an unmanned aerial vehicle (UAV) to capture RGB images of rice fields.
• Proposed RiceNet, a deep learning model with a feature extractor and three decoders (density map, location, size).
• Incorporated a rice plant attention mechanism and positive-negative loss for improved accuracy.

Main Results:

• Developed a new UAV-based rice counting dataset with 355 images and 257,793 labeled points.
• RiceNet achieved mean absolute error (MAE) of 8.6 and root mean square error (RMSE) of 11.2.
• Demonstrated superior performance compared to state-of-the-art methods on three different crop datasets.

Conclusions:

• RiceNet offers an accurate and efficient alternative to manual rice plant counting.
• The proposed method has the potential to revolutionize rice production management.
• Automated plant counting using UAVs and deep learning is a viable approach for agricultural applications.