UAV-based individual Chinese cabbage weight prediction using multi-temporal data.

Area of Science:

• Agricultural Science
• Remote Sensing
• Computer Vision

Background:

• Unmanned aerial vehicles (UAVs) are increasingly used for crop monitoring and yield prediction.
• Predicting individual plant harvest weight using UAV data is challenging due to difficulties in extracting plant-specific features.
• Existing methods often struggle with individual-level analysis and early prediction.

Purpose of the Study:

• To develop and validate a method for automatic detection and extraction of multitemporal individual plant features from UAV data.
• To predict the individual harvest weight of Chinese cabbage plants using these extracted features.
• To assess the feasibility of early harvest weight prediction using UAV-based data.

Main Methods:

• Acquired RGB and multi-spectral imagery from UAVs over 1196 Chinese cabbage plants.
• Utilized an object detection algorithm on RGB orthomosaic images for individual plant detection (>95% accuracy).
• Applied feature selection and analyzed multitemporal data resolutions to predict harvest weight using regression models.

Main Results:

• Achieved a high coefficient of determination (R² = 0.86) and low root mean square error (RMSE = 436 g/plant) for harvest weight prediction.
• Demonstrated successful prediction (R² > 0.72, RMSE < 560 g/plant) up to 53 days before harvest.
• Validated the effectiveness of multitemporal features for predicting individual plant weight.

Conclusions:

• The proposed method accurately predicts individual Chinese cabbage harvest weight using UAV-derived data.
• Multitemporal feature analysis significantly enhances the capability for early and accurate yield prediction.
• This approach offers a valuable tool for precision agriculture and improved crop management.