WINTER WHEAT YIELD ESTIMATION WITH GROUND BASED SPECTRAL INFORMATION

Timely and accurate knowledge of crop yield information is of great significance for guiding agricultural production, formulating agricultural policies and controlling the balance of food supply-demand. In the process of yield formation, spectral analysis technology, as a fast and non-destructive method, has been widely applied to detect the photosynthetic capacity indexes, which is highly correlative with the final crop yield. In this study, the spectral information of winter wheat leaves was obtained in three important phenological phases. The two-dimensional (2D) correlation spectrum analysis modified by the mutual information (MI) was then brought in, and the dynamic spectra were got by using crop yields as the perturbation quantity. After the analysis of the photosynthesis mechanism and 2D synchronous correlation spectra in three periods, three groups of wavebands were selected as the sensitive spectral information to the final yields. The principal component analysis (PCA) was conducted among the acquired wavebands. The crop yield was forecasted by the principal components from the selected wavelengths in different phenological phases. PLS model, the model based on neural network and the neural network prediction model optimized by the genetic algorithm were established separately. After the comparison, the BP Neural network model optimized by the genetic algorithm got a significant improvement in yield estimation accuracy. The results offered the rapid, convenient and valuable guidance for the agricultural production.