Monitoring Rice Growth Conditions Using UAV Remote Sensing System

The UAV remot sensing system provides a flexible and effective approach to acquire the environmental information of farmland,as well as the growth information of crop.It has been applied in agricultural production and scientific research in recent years.With the advent of the agricultural 4.0 era,thte UAV remote sensing has become an important part of smart agriculture,providing data support and decision-making basis for smart agricultural management.This research takes rice and other crops as the research object,and uses drones equipped with different types of sensors such as hyperspectral,multispectral and visible light to obtain low-altitude remote sensing images of crops.Combining field survey sampling data,spectral analysis,image analysis,photogrammetry,statistical analysis and other technologies were used to study the theory and methods of agronomic parameters estimation,including nutrient content,chlorophyll,leaf area in dex,plant height,as well as rapid and quantitative monitoring technology for agricultural disasters suh as lodging,insect pests,etc.This research aims to explore the application of drones in crop growth information acquisition and growth monitoring.The main conclusions of this research are listed as follows:?1?UAVs equipped with hyperspectral imagers were used to obtain hyperspectral images of rice canopy and diagnose the content of nitrogen,phosphorus and potassium in rice canopy leaves.The results showed that the spectral characteristics of rice canopy leaf nitrogen content?LNC?,leaf phosphorus content?LPC?and leaf potassium content?LKC?are consistent.The contents of the three nutrient elements were significantly negatively correlated with the spectral reflectance of the rice canopy on the UAV image in the 462?718 nm wavelength range?P<0.001?and were negatively correlated with the first derivative spectrum significantly in the visible light range of478?626nm and the near-infrared range of 782?886nm?P<0.001?.Using the spectrum value of the characteristic wavelengths selected by the continuous projection algorithm as the independent variable,the estimation model of the content of three elements in the rice canopy were built.The validation R²of these models were all above 0.8.The LNC was significantly related to the NDSI(R₅₂₆,R₅₆₂),RSI(R₅₂₆,R₅₆₂),DSI(R₅₈₂,R₅₀₂),NDSI(D₅₄₂,D₆₆₆),RSI(D₅₈₂,D₆₅₄)and DSI(D₅₅₄,D₆₄₆).The LPC was significantly related to the NDSI(R₄₉₈,R₆₀₆),RSI(R₄₉₈,R₆₀₆),DSI(R₄₉₈,R₅₈₆)?NDSI(D₆₄₂,D₆₅₀),RSI(D₆₅₀,D₈₃₈),DSI(D₆₁₄,D₆₄₆).The LKC was significantly related to the NDSI(R₅₁₄,R₅₇₀)?RSI(R₅₁₄,R₅₇₀)?DSI(R₄₉₈,R₅₈₂)?NDSI(D₆₃₈,D₆₅₄),RSI(D₆₄₂,D₆₅₀)?DSI(D₆₁₈,D₆₄₂).The correlation coefficients were all above 0.85.The models based on the newly-built spectral index have good predictive ability for LPC with validation R² higher than 0.8.The spatial distribution of LNC,LPC and LKC in the rice canopy at each growth stage was calculated based on the model and hyperspectral images.The results are consistent with the measured values and can be used to monitor the abundance and deficiency of nitrogen,phosphorus and potassium in the rice canopy leaves.?2?The spectral imageries of two paddy rice fields with different cultivars in Ningxia?NX?and Shanghai?SH?were acquired using the hyperspectral imager?Cubert S185?and multispectral imager?Micasense Red Edge 3?mounted on the UAV platforms.At the same time,the canopy leaf SPAD values were measured using a SPAD-502 Chlorophyll Meter.The spectral features of the canopy leaf SPAD values in each study area were analyzed and the results showed that the rice canopy leaf SPAD values in both regions had significant negative correlations with the 560nm?green?,668nm?red?and 840nm?NIR?band at 0.01 level.The absolute value of the maximum correlation coefficient reached 0.88.SPAD values were also significantly correlated to 8 vegetation indices at 0.01 level.Among these vegetation indices,only NPCI had a negative correlation with the rice canopy leaf SPAD values,while the correlations of the others were positive,and the absolute values of the correlation coefficient are within the range of 0.4?0.85.By taking the reflectance of sensitive bands and the vegetation indices as independent variables,the SPAD value estimation models for each group were built respectively using multiple regression methods,which was based on the machine learning theory.The performance of SVR model achieved the best.The R² of the SVR model for NX-SH was 0.84 and the root mean square error was 2.93.The results demonstrated that the rice canopy leaf SPAD values in different regions,cultivars and different types of sensor-based data shared similar spectral features,and could be estimated by universal models.This research could provide the theoretical basis for the crop growth monitoring in different agroclimatic regions using UAV remote sensing.?3?The spectral features of rice and wheat leaf chlorophyll content?LCC?and leaf area index?LAI?on the UAV hyperspectral images were analyzed.The results showed that the correlation between rice and wheat LCC and spectral reflectance showed a relatively stable and significant negative correlation in the visible light region?P<0.01?.The LAI of both crops and the spectral reflectance in the NIR band were strongly correlated positively?P<0.01?.The LCC-LAI collaborative model were constructed using the partial least squares regression method,which was able to calculate the multiple dependent variables from multiple independent variables.When the internal autocorrelation of the independent variables was high,the estimation accuracy of the collaborative model on LCC and LAI is higher than that of the univariate model,indicating that the multi-dependent variable collaborative algorithm can improve the model's ability to predict LCC and LAI of rice and wheat.In addition,a general model of rice+wheat LCC-LAI was constructed,and the predicted R² of LCC and LAI of the two crops reached above0.65,indicating that the general model could estimate the LCC and LAI of the two crops in the scenario of simultaneously monitoring wheat and rice.?4?The high-precision surveying and mapping drones were used to obtain multi-temporal digital surface models of rice breeding plots.Through the analysis of DSM in different periods,the plant height information of rice in different periods was extracted,and the plant height value of each plot in each period is calculated,and the spatial distribution of plant height is obtained.The plant heights extracted from DSM were verified using ground measurements.The verified R² in each period was higher than 0.7,the RMSE was lower than 0.07,and the maximum error was no more than 0.1m.?5?Multispectral and RGB?red,green,and blue?optical cameras mounted on UAV?unmanned aerial vehicle?platforms were used to acquire images of a rice paddy.Image features of non-lodged and lodged rice,including spectral reflectance,vegetation indices,texture,and colour,were extracted and analysed in order to optimize the indicators for lodging detection.Rice lodging detection models based on the selected image features were built to discriminate between non-lodged and lodged rice.Results revealed that the reflectance from the multispectral images at the green,red-edge and NIR bands were the optimum indicators of lodged rice.Moreover,for the RGB imagery,the Mean?G,Variance?B,g and Ex G4 were determined as the most optimal rice lodging indicators.The proposed models demonstrated lodging detection accuracies that reached over 90%,with the multispectral lodging detection model proving to be more accurate than the corresponding RGB model.Our results demonstrate that UAV-based remote sensing can play an important role in assessing the damage of rice lodging.?6?By taking the rice leaf roller as the research object,the multispectral images of rice field were acquired by a UAV-based remote sensing system and the rate of roll leaf was investigated.Then the spectral and texture features of rice in different infestation levels were analyzed.Based on the sensitive spectral and texture features,the estimation models of the rate of roll leaf were established to evaluate the infestation levels.The results showed that the rate of roll leaf had significant negative correlation with the spectral reflectance at green?475nm?,red-edge?717nm?and NIR?840nm?bands and positive correlation with the spectral reflectance at red band.The rate of roll leaf was also negatively correlated with NDVI and DVI.Leaf roll rate was significantly correlated with 4 texture features,such as mean,homogeneity,contrast and dissimilarity in green red,red edge and NIR bands.The estimation models of the rate of roll leaf established based on both spectral and texture features by ANN achieved the best accuracy.The validation R² was 0.717 and the RMSE was 0.702.The distribution map of the rate of roll leaf based on the PLSR model was consistent with the field survey.The results could be used as a method for the rapid investigation and precise control of insect infestation.