Evaluation of important phenotypic parameters of tea plantations using multi-source remote sensing data
Frontiers in Plant Science
|August 8, 2022
Summary
Unmanned aerial vehicles (UAVs) with multiple sensors effectively monitor tea plant health. Combining data from various sensors significantly improves the accuracy of assessing tea canopy parameters like height and leaf area index.
Area of Science:
- Agricultural remote sensing
- Plant phenotyping
- Machine learning applications in agriculture
Background:
- Accurate monitoring of tea plant growth parameters is crucial for plantation management.
- Traditional methods are labor-intensive and lack efficiency for large-scale assessment.
- Unmanned aerial vehicle (UAV) multi-source remote sensing offers a promising solution for enhanced crop monitoring.
Purpose of the Study:
- To efficiently monitor key phenotypic parameters of tea canopies using UAV-based remote sensing.
- To evaluate the performance of single-source versus multi-source data with various machine learning algorithms.
- To identify optimal sensor combinations and models for assessing tea plant health indicators.
Main Methods:
- Deployment of UAVs equipped with multispectral, thermal infrared, RGB, LiDAR, and tilt photography sensors.
- Acquisition of comprehensive remote sensing data for tea canopies.
- Application of four machine learning algorithms (including SVM and RF) to model single-source and multi-source data for estimating height (H), leaf area index (LAI), canopy water content (W), leaf chlorophyll content (LCC), and leaf nitrogen concentration (LNC).
Main Results:
- Multi-source data significantly improved the accuracy and robustness for estimating H, LAI, W, and LCC.
- Optimal estimations were achieved with LiDAR + TC for H (SVM model), LiDAR + TC + MS for LAI (SVM model), RGB + TM for W (SVM model), and MS + RGB for LCC (RF model).
- Single-source multispectral (MS) data with the RF model provided the best estimation for LNC.
Conclusions:
- UAV multi-source remote sensing provides an effective high-throughput technique for tea crown phenotypic information.
- Specific sensor combinations and machine learning models are recommended for accurate assessment of different tea phenotypic parameters.
- This study offers a guiding principle for utilizing artificial intelligence in tea plantation management.
