Study On The Scaling And The Establishion Of The Hyper-spectral Inversion Model Of Chllorophyll Content In Rice

Chlorophyll is the main pigment of photosynthesis in crops. It directly controls the cycles of matter and the transmission of energy. So that the change of its content can be used to evaluate the ability of crop photosynthesis, the stress levels of heavy-metal pollution as well as the evaluation of the crop’s nutrient levels. Therefore, it is important for the agricultural production to monitor crop chlorophyll content.In this study, we select four rice fields of Changchun as the sampling area. Based on ASD spectral data and rice chlorophyll content measured in the sample area, we establish the hyper-spectral inversion model of chlorophyll content in rice. On the basis of the shape of “peak and valley” in the spectral response curves of plant, we apply the established model in large area which is scaled up using the Hyperion data. The main work and conclusions are as follows:i. In this study, we chose the former model established using BP neural network model as the hyper-spectral inversion model of chlorophyll content in rice. The input parameters of model are four vegetation indices based on results of previous studies and experimental analysis. The model contained a hidden layer and the structure is 4-12-1. The output parameters are the data of chlorophyll content. As a result, the correlation coefficient R2 = 0.882, RMSE = 2.958.ii. Based on the analysis of the established spectral inversion models of chlorophyll content in rice, through the comparison of the spectral inversion models established based on artificial neural network BP and other statistical models which were established using the original spectrum, the deformation of the original spectrum and their first derivative, we found that the relevant factor has greatly improved, meanwhile, the root mean square error has lower than the statistical models. So that we can know that the former model is better to monitor the chlorophyll content of rice.iii. Through research, we found that the spectral curve of rice has significant "peak-valley" morphology. ASD spectrum data and pixel spectrum data can be represented as a piecewise function that the wavelength λ is as independent variable, and the reflectance data as dependent variable, and the two curves are similar in shape. By the mathematical knowledge, we can know that the two curves can be converted through the flex and pan of curve.iv. Through the "peak- valley" morphology in the spectrum, we establish linear conversion relationship between ASD spectral reflectance data and pixel reflectance data, in order to scale up the established model. Through it, we can monitor the chlorophyll content of rice in a large area achieving fast, non-destructive remote sensing of chlorophyll content using Hyperion remote sensing images.