The Normalized Difference Vegetation Index (NDVI) is an important remote sensing index that is widely used to assess vegetation coverage, monitor crop growth, and predict yields. Traditional NDVI calculation methods often rely on multispectral or hyperspectral imagery, which are costly and complex to operate, thus limiting their applicability in small-scale farms and developing countries. To address these limitations, this study proposes an NDVI estimation method based on low-cost RGB (red, green, and blue) UAV (unmanned aerial vehicle) imagery combined with deep learning techniques. This study utilizes field data from five major crops (cotton, rice, maize, rape, and wheat) throughout their whole growth periods. RGB images were used to extract conventional features, including color indices (CIs), texture features (TFs), and vegetation coverage, while convolutional features (CFs) were extracted using the deep learning network ResNet50 to optimize the model. The results indicate that the model, optimized with CFs, significantly enhanced NDVI estimation accuracy. Specifically, the R² values for maize, rape, and wheat during their whole growth periods reached 0.99, while those for rice and cotton were 0.96 and 0.93, respectively. Notably, the accuracy improvement in later growth periods was most pronounced for cotton and maize, with average R² increases of 0.15 and 0.14, respectively, whereas wheat exhibited a more modest improvement of only 0.04. This method leverages deep learning to capture structural changes in crop populations, optimizing conventional image features and improving NDVI estimation accuracy. This study presents an NDVI estimation approach applicable to the whole growth period of common crops, particularly those with significant population variations, and provides a valuable reference for estimating other vegetation indices using low-cost UAV-acquired RGB images.

Leaf nitrogen content (LNC) is a vital agronomic parameter in rice, commonly used to evaluate photosynthetic capacity and diagnose crop nutrient levels. Nitrogen deficiency can significantly reduce yield, underscoring the importance of accurate LNC estimation for practical applications. This study utilizes hyperspectral UAV imagery to acquire rice canopy data, applying various machine learning regression algorithms (MLR) to develop an LNC estimation model and create a nitrogen concentration distribution map, offering valuable guidance for subsequent field nitrogen management. The analysis incorporates four types of spectral data extracted throughout the rice growth cycle: original reflectance bands (OR bands), vegetation indices (VIs), first-derivative spectral bands (FD bands), and hyperspectral variable parameters (HSPs) as model inputs, while measured nitrogen concentration serves as the output. Results demonstrate that the random forest regression (RFR) and gradient boosting decision tree (GBDT) algorithms performed effectively, with the GBDT achieving the highest average R² of 0.76 across different nitrogen treatments. Among the nitrogen estimation models for various rice varieties, RFR exhibited superior accuracy, achieving an R² of 0.95 for the SuXiangJing100 variety, while the GBDT reached 0.93. Meanwhile, the support vector machine regression (SVMR) showed slightly lower accuracy, and partial least-squares regression (PLSR) was the least effective. This study developed an LNC estimation method applicable to the whole growth stage of common rice varieties. The method is suitable for estimating rice LNC across different growth stages, varieties, and nitrogen treatments, and it also provides a reference for nitrogen estimation and fertilization planning at flight altitudes other than the 120 m used in this study.

Real-time monitoring of rice-wheat rotation areas is crucial for improving agricultural productivity and ensuring the overall yield of rice and wheat. However, the current monitoring methods mainly rely on manual recording and observation, leading to low monitoring efficiency. This study addresses the challenges of monitoring agricultural progress and the time-consuming and labor-intensive nature of the monitoring process. By integrating Unmanned aerial vehicle (UAV) image analysis technology and deep learning techniques, we proposed a method for precise monitoring of agricultural progress in rice-wheat rotation areas. The proposed method was initially used to extract color, texture, and convolutional features from RGB images for model construction. Then, redundant features were removed through feature correlation analysis. Additionally, activation layer features suitable for agricultural progress classification were proposed using the deep learning framework, enhancing classification accuracy. The results showed that the classification accuracies obtained by combining Color+Texture, Color+L08CON, Color+ResNet50, and Color+Texture+L08CON with the random forest model were 0.91, 0.99, 0.98, and 0.99, respectively. In contrast, the model using only color features had an accuracy of 85.3%, which is significantly lower than that of the multi-feature combination models. Color feature extraction took the shortest processing time (0.19 s) for a single image. The proposed Color+L08CON method achieved high accuracy with a processing time of 1.25 s, much faster than directly using deep learning models. This method effectively meets the need for real-time monitoring of agricultural progress.

Leaf water content (LWC) is a key physiological parameter for assessing maize moisture status, with direct implications for crop growth and yield. Accurate LWC estimation is essential for water resource management and precision agriculture. This study introduces a high-precision method for estimating maize LWC utilizing UAV-based multispectral imagery combined with a Random Forest Regression (RFR) model. By extracting vegetation indices, image coverage, and texture features and integrating them with ground-truth data, the study examines the variation in LWC estimation accuracy across different growth stages. The results indicate that the RFR model performs optimally during the seedling stage, with a root relative mean square error (RRMSE) of 2.99%, whereas estimation errors are larger during the tasseling stage, with an RRMSE of 4.13%. Moreover, the RFR model consistently outperforms multiple linear regression (MLR) and ridge regression (RR) models throughout the growing season, demonstrating lower errors on both training and testing datasets. Notably, the RFR model exhibits significantly reduced errors in the training dataset compared to both MLR and RR models. Following particle swarm optimization (PSO), the prediction accuracy of the RFR model is notably enhanced, with the RRMSE on the training dataset decreasing from 1.46% to 1.19%. This study provides an effective approach for estimating maize LWC across different growth stages, supporting crop water management and precision agriculture, and offering valuable insights for the estimation of water content in other crops.