Research On The Estimation Of Citrus Yield Based On The Spectral Data And Agronomy Parameters

In recent years, remote sensing has been widely applied in crop yield estimation and become the focus of research and attention. It is also a new technology developed in recent years. Hyperspectral data has advantages, such as objective, rapid, large amount of information and so on, and also can provide more precise spectral information. Studies have shown the precise spectral information have a good relationship with vegetation index, chlorophyll content and moisture content, etc. Therefore, it is feasible that we build models by measuring the spectral information of crop, vegetation index, chlorophyll content, moisture content and the yield of crop in order to achieve the yield estimation and predict. At the same time, it has the important practical significance on the production, management, yield estimation, and pest control of citrus.Taking the orange trees of a farm orchard in Xiema Town, Beibei District of Chongqing as the research object and using the spectral data of all samples in2010and2011measured by ASD handheld spectrometer, this article processes the spectral data, extracts the NDVI of citrus, and then combines the correlation of yield and NDVI. We know that the NDVI value extracted in May is the best NDVI for predicting yield. On the other hand, we obtain the agronomic parameters (N, P, K, moisture content, chlorophyll) of samples through the acquisition of citrus leaf, and make sure the best agronomic parameters of Citrus are chlorophyll content and nitrogen content according to the correlation with sample yield. Finally, we build the multivariate linear regression estimation model based on the normalized vegetation index (NDVI), nitrogen content, chlorophyll content and yield and do the significance test and sample test for the model. Through the significance test, we get the model correlation coefficient R is0.631, F is13.201and P is0.0001. This shows that the model has statistical significance and can be used for the actual yield estimation. By the sample test we get that the model prediction accuracy error is less than20%. It can satisfy the demand of practical estimation and has realistic directive significance.