Comparison RGB Digital Camera with Active Canopy Sensor Based on UAV for Rice Nitrogen Status Monitoring

Nitrogen (N) plays a vital role in crop growth and makes a great contribution to world food production. Aerial visible light (RGB) imagery acquired by consumer UAV-mounted digital camera is attracting more and more researchers' attention for crop remote sensing. Because of the suitability for different environmental conditions and convenience in data processing and analysis, active optical canopy sensors have been widely used for N-status monitoring. Accordingly, the system of UAV-based active sensor is expected to take both advantages of active sensors and U A V s. The objectives of this study were: (1) to compare RGB digital camera with active canopy sensor based on UA V platform for estimating rice N status; (2) to explore the method of multiple regression prediction (Random Forest) and the potential of combining data from UAV-based RGB imagery and active sensor for N status monitoring. Field experiment on rice involving 4 N rates (0, 135, 270 and 405 kg/ha), 2 transplanting ways and 2 varieties was conducted in 2017 at Lianyungang experiment station of Nanjing Agricultural University in Jiangsu Province. Plant sampling and sensing data collection (from aerial imagery and active sensor) were conducted at 4 key growth stages. LAI (leaf area index), LDM (leaf dry matter) and LNA (leaf area accumulation) were tested and calculated as the N-status indicators. Aerial images were acquired by a built-in RGB camera from a multi-rotor UA V named DJI Phantom 3 Professional. Spectral reflectance of red (670 nm), red-edge (730 nm) and near infrared (780 nm) wavebands were collected by a UA V -mounted active canopy sensor RapidSCAN CS-45. Vegetation indexes (VIs, calculated from active sensor reflectance) and color indexes (CI, calculated from digital number values of aerial images) were used to build predictive models for each N-status indicator respectively. Random forest estimator (RF) was used to explore the potential of multiple regression-based prediction and combining data from RGB imagery and active sensor together for rice leaf N -status monitoring. The results indicated great applicability of active canopy sensor on rice N -status monitoring compared to traditional RGB digital camera, and greater performance was shown by random forest estimator with mixed inputs of VIs and CIs together. Future studies are needed on practical evaluation and better multiple regression utilization for these two sensing systems.

I. Introduction

Nitrogen (N) plays a critical role in crop growth and makes a great contribution to world food production. For the demand of precision N management, it is crucial to develop high-efficient, reliable and usable methods for monitoring crop N status. Unmanned aerial vehicle(UA V) is a low-cost remote sensing platform which has great flexibility and mobility and attracts researchers' attention in the field of crop N and plant phenotyping estimation [1], [2].