| The aboveground fresh biomass is an important index for crop growth. The management of crop biomass is one of the most important factors of crop breeding and crop management, which can be the basic factor in detecting and estimating the yield of crop. The present single testing technique is also difficult to achieve accurate evaluation of biomass with the phenomena of biomass saturation. In this work, for the requirements of nondestructive and fast detecting, the spectroscopy technology, the machine vision technology, the mechanics characteristics of the stalk and the data fusion of different technology were used to construct the biomass prediction models, which achieved the comprehensive accurate evaluation of the wheat biomass. The main points are summarized as follows:(1) The biomass prediction models based on the mechanical property of wheat stalk were proposed. To study the adaptability and the accuracy, the prediction models of the different growth stages were tested. The detection system based on mechanical property of wheat stalk was developed to study the feasibility of method. The biomass prediction models of different growth stages were constructed. Experiments indicated that as the measuring site rising, the voltage signals of the mechanical property of wheat stalk were reduced. The voltage signals of the mechanical property of different measuring site had good correlation on wheat biomass. The determination coefficients(R2) were 0.683, 0.622, and 0.561 for measuring sites of 40 cm, 45 cm, 50 cm respectively. The measurement accuracy of biomass prediction models of different growth stages were different, of which the heading stage model had the best performance with R 2 =0.675 and RMSE=0.218 kg in calibration model, R 2 =0.623 and RMSE=0.216 kg in validation model.(2) The biomass prediction models based on image technology was constructed. The wheat coverage of canopy image in later growth stage approached 100 percent, which wasn’t related to biomass. The side images of samples at medium and later growth stage were collected to construct the calibration models by multivariate regression analysis. The predicting ability of the models was tested by the unknown samples. Experiments indicate that the wheat coverage changed with the variations of biomass during seedling period. The model of seeding stage had the best performance with the R2=0.675 and RMSE=0.851 kg in calibration model, R2=0.828 and RMSE=0.017 kg in validation model. With the changes of growth stage, the biomass had the saturation at the blooming and grouting stage, the accuracy of prediction model decreased gradually, which illustrated that the biomass prediction models based on image technology present limitations at later growth stage.(3) The biomass prediction models based on spectroscopy technology was constructed. Characteristic parameters of wheat canopy spectral were extracted, such as the spectral reflectance, spectral indices, red edge amplitude and characteristics wavelengths. The calibration models were constructed using PLS with the characteristic parameters and biomass. The predicting ability of the models was tested by the unknown samples. Experiments indicate that the model built with spectral reflectance in the 400-1000 nm had the best performance. For different stage, the model of seedling period had the best performance with the R2=0.839 and RMSE=0.016 kg in calibration model, R2=0.823 and RMSE=0.017 kg in validation model.With the changes of growth stage, the biomass had the saturation at the blooming and grouting stage, the accuracy of prediction model decreased gradually, which illustrated that the biomass prediction models based on spectroscopy technology present limitations at later growth stage.(4) Based on single technique for detecting the wheat biomass, the problems for the saturation and low-precision of the single technique models. Multiple information fusion technology by integrating spectroscopy technology, the machine vision technology and the mechanics characteristics of the stalk was used to build the models at different stages. Different feature variables were extracted and the data were fused in the feature layers as a new characteristic variable to construct the biomass prediction models of different stages. The experimental results showed that the model based on data fusion of two arbitrary technologies is better than that based on the single technique models. For the heading period, flowering period and filling period, the model based on data fusion of three technologies were better than the model from single sensors or two sensors information. To a certain extent, the problems for the saturation and the low-precision of the single technique models were improved. |
