Multispectral reflectance inversion and chlorophyll content diagnosis of maize at seeding stage

In order to explore the fast and non-destructive estimation method of chlorophyll content of maize, a fast and nondestructive diagnostic study of chlorophyll content index at the seedling stage of maize was carried out in this paper based on the 2-CCD multi-spectral image acquisition system. In the calibration experiment, the spectral curves of the 4 stage diffuse gray board and the standard white board was acquired by ASD's FieldSpec Handheld spectrometer firstly. The average values of reflectivity were calculated respectively in the four bands of 5(450~490 nm), G(530~570 nm), R(600~640 nm) and NIR (780~820 nm). The linear inversion formula between the normalized mean gray value and the spectral reflectance of the maize leaves in field was established by the normalized mean gray value of the gray board's multi-spectral images in the four bands. In order to eliminate the effect of illumination changes on the gray value of crop canopy image in the field experiment and convert the canopy image data of maize to the data of its leaf reflectance under different illumination conditions, calibration was carried out in real time by using the diffuse reflection gray board which have four different gray levels and meet the conditions of Lambertian in the field. 64 groups multispectral images of maize seedling canopy and diffuse reflectance gray board were acquired synchronously. Correlation regression analysis was carried out on the normalized average gray value of the diffuse reflectance gray board's multispectral images in the field experiment and the calibration experiment. According to the changing trend of the sunlight, the gray value of the maize canopy was corrected in each of the 8 samples. Average value of each sample correction factor was used as the coefficient of the correction formula. After the multi-spectral images of maize in seedling stage processing, the normalized average gray value of R, G, B and NIR was extracted from canopy image. And 4 common image vegetation indices (ANDVI, ANDCI, ARVI and ADVI) were calculated. The normalized average gray value was corrected by the correction formula of gray value, the average reflectivity of four bands were obtained by the inversion of linear formula. And 4 common spectral vegetation indices (RNDVI, RNDCI, RRVI and RDVI) were calculated. Correlation analysis between the parameters of the detection and the SPAD value of the chlorophyll content index was carried out before and after the correction. The results showed that: compared with the former, the correlation between the average reflectivity, the spectral vegetation index and SPAD value was significantly increased. The correlation coefficient of vegetation index was promoted from the low correlation (r<0.5) to significantly related (r>0.5). The fitting model of vegetation index (RNDVI, RRVI and RDVI) and chlorophyll content index was established by least squares support vector regression (LS-SVR). The results showed that the fitting determination coefficient was up to 0.73, and the fitting result was ideal. It is proved that the method was feasible to establish reflectance inversion correction model of maize canopy multi-spectral image by diffuse reflectance gray board. This method provided a support for the rapid and non-destructive diagnosis of chlorophyll content at maize seedling stage. © 2015, Chinese Society of Agricultural Engineering. All right reserved.