Field Crop 2025, Vol.8, No.4, 204-212 http://cropscipublisher.com/index.php/fc 206 such as red border and near-infrared, and can calculate vegetation indices (such as NDVI) for monitoring growth status, nitrogen level and stress conditions (Kumar et al., 2024). Hyperspectral camera: It can provide continuous information of hundreds of bands and is suitable for analyzing traits such as chlorophyll, leaf area index (LAI), and aboveground biomass (Yu et al., 2017; Ban et al., 2022). Thermal infrared camera: Capable of measuring canopy temperature, used to assess moisture conditions, evapotranspiration and irrigation requirements (Wu et al., 2025). The combined use of RGB, multispectral, hyperspectral and thermal imaging can improve the accuracy and stability of rice growth monitoring and stress detection (Xu et al., 2022; Shen et al., 2024; Guo et al., 2025). 2.3 Role of multi-temporal and high-resolution imagery in monitoring rice growth Obtaining drone images at different growth stages of rice can dynamically track the growth situation. This can also help detect stress earlier and assist in predicting production. High-resolution images can display small-scale differences in the field, facilitating targeted management. If multi-temporal images are combined with machine learning, leaf area index, biomass and yield can be estimated more accurately. Meanwhile, this method can also provide real-time decision support for fertilization, irrigation and pest control (Luo et al., 2022; Chen et al., 2024; Gade et al., 2024). 3 Remote Sensing Characterization of Key Rice Growth Indicators 3.1 Vegetation indices (NDVI, EVI, PRI, etc.) and rice canopy parameters Vegetation indices (VI) such as NDVI, EVI and PRI can be calculated from multispectral, hyperspectral or RGB images collected by drones. These indices are often used to estimate the canopy parameters of rice, including leaf area index (LAI), chlorophyll content (SPAD), biomass and nitrogen level. Research has found that some indices, especially those containing red edges and near-infrared bands, have a high correlation with LAI and SPAD at different reproductive stages (Zha et al., 2020; Ban et al., 2022). For instance, mND705, SAVI and WDRVI perform well in estimating LAI, while indices such as GLI, RGRI and ExR are more suitable for reflecting biomass and nitrogen content (Prabhakar et al., 2024). If VI is combined with texture or structural features, the monitoring accuracy can be further improved (Lyu, 2024). 3.2 Structural parameters extracted from UAV imagery (plant height, canopy coverage, LAI) Drone images can directly obtain some structural parameters, such as plant height, canopy coverage and leaf area index (LAI). These indicators are closely related to the growth and yield of rice. Plant height can be calculated from RGB or multispectral images through digital surface models. Coverage and LAI can be estimated by spectral and texture analysis (Duan et al., 2019; Liao et al., 2025). Studies have found that if canopy height and VI are combined, the estimation of LAI will be more accurate, have less error, and can also reduce the impact of phenological changes (Gong et al., 2021). High-resolution unmanned aerial vehicle images can also draw the distribution of these parameters in the field, providing a reference for precise management (Qiu et al., 2020). 3.3 Remote sensing inversion methods for rice physiological and ecological indicators Nowadays, many studies have begun to use machine learning methods, such as random forests, support vector regression and neural networks, to invert the physiological and ecological indicators of rice. These methods, combining spectral, texture and structural features, can predict LAI, SPAD, biomass, nitrogen nutrient index (NNI) and yield relatively accurately (Liu et al., 2023; Wang et al., 2023). If the data from drones and satellites are combined and deep learning models are used, the prediction accuracy and spatial details will be better. Relatively stable monitoring can be achieved even when the field conditions are very complex (Chen et al., 2024; Li et al., 2024). Therefore, the combination of multi-source and multi-temporal data with advanced modeling is the key to achieving high-precision and large-scale rice monitoring (Li and Jiong, 2024). 4 Applications of Machine Learning in Rice Growth Monitoring 4.1 Suitability analysis of common algorithms (RF, SVM, XGBoost, deep learning) In rice growth monitoring, machine learning (ML) is often used to analyze the data collected by drones. Among traditional methods, random Forest (RF) and Support Vector Machine (SVM) are the most common. They can handle nonlinear relationships and the results are relatively stable. Research has found that RF is often more accurate than linear regression and other models in estimating nitrogen status, biomass and growth stage (Zha et
RkJQdWJsaXNoZXIy MjQ4ODYzNA==