Prediction Of Wheat Yield Reduction Under Drought Stress Based On UAV Hyperspectral Remote Sensing Technology

Crop disaster caused by drought is a worldwide agrometeorological disaster,which may occur in all stages of wheat growth and development,leading to wheat plant water loss and wheat yield reduction to a certain extent.The yield reduction rate of wheat is a good indicator of the impact of drought stress on crops.UAV hyperspectral remote sensing technology provides a feasible method to identify the degree of crop affected by drought stress.Therefore,the study of wheat yield reduction rate under drought stress based on UAV hyperspectral remote sensing image has important practical application value and significance for disaster monitoring and production management decision-making.Based on the experimental site of dry land water saving agriculture in Hebei province as the research area,using water control experiment field environment,for the wheat canopy UAV under different water treatment hyperspectral remote sensing image data,ground wheat yield data,etc.,and combined with a variety of spectral image data processing and data analysis,modeling technology,explore in different key stages,The spectral characteristics of wheat canopy under different degrees of drought stress were selected and the sensitive spectral indexes were selected to reflect the degree of yield loss and drought stress.Based on the selected sensitive spectral index,a prediction model of wheat yield reduction was established to evaluate the accuracy of the selected spectral index and the model in predicting wheat yield reduction at the field scale.The main research contents are as follows:(1)For UAV load noise spectral reflectance in hyperspectral image,based on the principle of spectral reflectance errors,use S＿G filtering smoothing algorithm,at the same time introduced the error weighting index (24)and fitting effect 6),in order to achieve adaptive noise removal under different spectral characteristics of different parameters,and finally to the whole image is processed in the study area,The results show that the adaptive smoothing not only removes the spectral noise,but also preserves the spectral characteristic information well,improves the image quality and application ability,and provides a basis for the follow-up research(2)Based on the spectral data of wheat yield reduction and canopy reflectance,correlation analysis was used to explore the relationship between spectral reflectance and wheat yield reduction.The results showed that there was significant correlation between the reflectance of wheat canopy spectrum at different wavelengths and wheat yield reduction.Based on the data of wheat canopy spectral reflectance and wheat yield reduction rate,26 existing narrow-band spectral indices and 9 combined narrow-band spectral indices were selected to better respond to wheat yield reduction rate during the whole growth period,which provided a theoretical basis for quantitative prediction of wheat yield reduction rate.(3)In different key growth periods of wheat,a linear regression prediction model and a partial least squares regression prediction model were constructed based on each optimal index and wheat yield reduction rate.In different key growth periods,the models constructed by each index showed relatively stable results in accuracy.Based on the partial least squares regression model of each index,when the principal components were 4,the model achieved the most stable results in each key growth stage of wheat.The correlation between broadband spectral index after spectral resampling and wheat yield reduction rate was analyzed,and the relevant conclusions were helpful to guide the quantitative study of crop physical and chemical parameters based on UAV hyperspectral.The key growth period of weighted based on continuous wheat production rate prediction model,and further discusses the accumulation effect of drought stress,reflect on the different development stages of the wheat under drought stress on wheat yield of cumulative impact,and the results show that based on continuous critical fertility weighted on the precision of prediction model has achieved more in line with the actual results.(4)Based on the principle of spectral feature change in response to crop physical and chemical parameters and the development and application of machine learning correlation models,the feasibility of BP neural network machine learning model based on feature band and deep learning model based on convolutional neural network in wheat yield and yield prediction was analyzed.By acquiring hyperspectral remote sensing image of the UAV data and output data on the ground,build a relevant model and carry on the training of the model and tuning parameters,the results show that based on machine learning related model of UAV hyperspectral remote sensing image in the prediction of wheat under drought stress on the research of production rate is feasible,provide reference for subsequent research...