Estimation of rice leaf nitrogen contents based on hyperspectral LIDAR

Precision agriculture has become a global research hotspot in recent years. Thus, a technique for rapidly monitoring a farmland in a large scale and for accurately monitoring the growing status of crops needs to be established. In this paper, a novel technique, i.e., hyperspectral LIDAR (HL) which worked based on wide spectrum emission and a 32-channel detector was introduced, and its potential in vegetation detection was then evaluated. These spectra collected by HL were used to classify and derive the nitrogen contents of rice under four different nitrogen content levels with support vector machine (SVM) regression. Meanwhile the wavelength selection and channel correction method for achieving high spectral resolution were discussed briefly. The analysis results show that: (1) the reflectance intensity of the selected characteristic wavelengths of HL system has high correlation with different nitrogen contents levels of rice. (2) By increasing the number of wavelengths in calculation, the classification accuracy is greatly improved (from 54% with 4 wavelengths to 83% with 32 wavelengths) and so the regression coefficient r2 is (from 0.51 with 4 wavelengths to 0.75 with 32 wavelengths). (3) Support vector machine (SVM) is a useful regression method for rice leaf nitrogen contents retrieval. These analysis results can help farmers to make fertilization strategies more accurately. The receiving channels and characteristic wavelengths of HL system can be flexibly selected according to different requirements and thus this system will be applied in other fields, such as geologic exploration and environmental monitoring.