Field Crop 2025, Vol.8, No.4, 204-212 http://cropscipublisher.com/index.php/fc 208 5.3 Pipeline design for UAV- and ML-based rice growth monitoring A typical unmanned aerial vehicle (UAV) + machine learning monitoring process roughly includes the following steps: Data collection: Obtain high-resolution images from unmanned aerial vehicles (UAVs) and collect auxiliary data such as weather, soil, and management (Chen et al., 2024). Preprocessing: Perform illumination correction, noise reduction, calibration and registration on the image. Feature engineering: Extract and screen relevant spectral, texture and structural features. Data Fusion: Integrating Multi-source Data to enrich Model Input (Du et al., 2024). Model training and validation: Trained using machine learning algorithms (such as RF, SVM or ensemble methods), and evaluated through cross-validation and independent test sets (Zha et al., 2020; Islam et al., 2023). Prediction and Mapping: Generate spatial distribution maps of rice growth to facilitate precise management (Chen et al., 2024; Shen et al., 2024). This process is modular, capable of supporting scalable, real-time and precise rice monitoring, and can also provide data support for precision agriculture. 6 Case Studies: Monitoring Practices in Large-scale Rice Fields 6.1 UAV monitoring cases in major rice production areas of the middle and lower Yangtze River Basin, China Unmanned aerial vehicle (UAV) remote sensing is a flexible and efficient technology that can obtain information on farmland environment and crop growth. In recent years, it has been increasingly applied in agricultural production and scientific research. In the research of the Yangtze River Basin, it was found that unmanned aerial vehicle remote sensing has a very good effect on the monitoring and management of rice. For instance, by combining the vegetation index obtained by drones with the depth image features and then analyzing them with a random forest model, it is possible to accurately identify the nutritional deficiencies of rice and formulate reasonable fertilization plans. The classification accuracy rate exceeds 96% (Figure 2) (Chen et al., 2025). In addition, combining drone data with satellite data can also enhance the accuracy of rice growth and pest monitoring. In the demonstration fields near Nanjing, this method is very effective for monitoring pests such as Cnaphalocrocis medinalis. It can also more accurately invert the leaf area index (LAI) and provide clearer spatial information, facilitating precise management (Chen et al., 2024). Figure 2 Field imagery acquisition and dataset preparation. (A) Specific parameters of the UAV camera and the field flight mission. (B) Acquisition and stitching of field imagery. (C) Calculation and merging of features as well as dataset preparation (Adopted from Chen et al., 2025) 6.2 Multi-temporal UAV monitoring and yield prediction studies in Southeast Asia In Southeast Asia, multi-temporal drone monitoring has been widely used to assess rice growth and predict yields. In southern China, researchers used drones equipped with RGB and multispectral cameras to track the entire rice-growing season. They extracted the vegetation index, canopy height and coverage, and combined them with the random forest model to successfully predict the yield. The coefficient of determination R² reaches 0.85 and can be applied across different years and plots (Wan et al., 2020). In Japan, researchers compared the effects of
RkJQdWJsaXNoZXIy MjQ4ODYzNA==