Low-Altitude Unmanned Aerial Vehicle-Based Remote Sensing For Rice Growth Status Monitoring

Remote sensing is a critical kind of technologies for quick and real-time estimation of crop growth and nitrogen status,which offers great support for real-time crop growth diagnosis and precision farming.Low-altitude unmanned aerial vehicle(UAV),as a new portable,convenient and easy-to-use aerial remote sensing platform,are gradually being used in the research of agricultural information monitoring.Rice field experiments were conducted in this study over different years,locations,varieties,nitrogen rates and transplanting methods to find out suitable UAV-based strategies for rice growth monitoring.For these two types of UAV-based remote sensing systems in this study,our targets were to provide scientific support and technical guidance about model construction and analysis,feature selection and comparison,technological innovation and optimization.In view of the of the UAV-based sensing with active canopy sensor,system construction method and appropriate operational mode were taken forward using the multi-rotor UAV DJI S1000+equipped with the active canopy sensor RapidSCAN CS-45.With the vegetation indices calculated by the red,red and near-infrared wavebands with RapidSCAN CS-45,the predictive models for dry matter weight(LDM),leaf area index(LAI)and leaf nitrogen accumulation(LNA)were constructed,respectively.The testing data obtained by handheld and UAV-based modes(under different sensing heights)were used to validate the models.The results show that the vegetation indices based on near-infrared and red-edge bands(NDRE and RERVI)have good predictive performance on each rice growth indicators,and R2 is 0.77,0.79 and 0.83,respectively.It is verified that the models are suitable for both handheld and UAV-based modes for practical utilization.The evaluation of the validation results also demonstrated that the flight height of 1.5 m above the canopy is suitable for the practical application.In view of the UAV-based sensing with digital camera,this study focuses on the application of a monitoring system for the rice leaf area index LAI using consumer-grade drone DJI Phantom 3 Professional and a built-in digital camera.In every key growth stage stages of rice,the UAV-based RGB digital images of the crop canopy were obtained,mosaiced and processed,and the DN values of the red,green and blue channels and the gray level co-occurrence matrix-based texture features(Texture)were extracted;then,commonly-used color indices(CI)and normalized texture indices(TI)were calculated out;the calibration results showed that VARI and NDTI(Rmea,Gmea)ranked the best in color indices and texture indices respectively,which have the optimum linear relationships with rice LAI.Different multivariate regression models,combined with three different input variables combinations(CI,Textures,CI+Textures),were established for rice LAI estimation.The validation results showed that the random forest can effectively avoid model over-fitting when combining different input variables,and construct a more robust rice LAI predictive model,which is suitable for practical use;with the inputs of color information(CI)and textures(Textures),the random forest method effectively improved the performance of the rice LAI estimation compared to the traditional method of using the color indices alone(R2=0.839,RMSE=0.874,MAE=0.694 in independent validation).This study is aimed at the urgent need in today’s crop production industry on real-time monitoring of crop growth,two types of UAV-based remote sensing modes for crop monitoring were evaluated;and these two modes showed their own pros and cons in data acquisition procedures,estimation capabilities and sensing scopes.However,further improvements are still needed in the issues like automatic processing of the remotely sensed spectrum and images,and the sensed data stabilization.In addition,according to the specific character of different crops,it is critical to combine the point data sensed from the UAV-based active sensors and the surface data from the digital camera for a universal strategy of crop growth monitoring,diagnosing and precision management in the future studies.