Research On Irrigation Water Requirement Inversion And Forecasting Based On Remote Sensing And Its System Implementation

The rational and optimal allocation of water resources for irrigation plays a vital role in improving the utilization rate and promoting the sustainable development of agricultural water resources.At present,the traditional way of irrigation is mostly used,which results in large quantities of water are wasted during irrigation process.The high-resolution remote sensing images can accurately monitor the change of each plot,and the use of deep semantic segmentation network can classify land cover to pixel-level.Based on remote sensing and deep learning technology,this thesis takes Xiaokaihe irrigation area of Shandong Province as an example to carry out the research on techniques for calculating and forecasting water requirement for irrigation,and mainly achieves the following research results.An accurate classification of crop cultivation is the basis for inversion of irrigation water requirements using remote sensing imagery.Based on deep semantic segmentation network,this thesis develops a classification model of planting structure and mix multiple exponents to participate in image fusion during preprocessing of remote sensing images,to enrich spectral complexity,improve spatial resolution,and enhance feature information.The depth-separable-convolution is used to replace the ordinary convolution,the Mish activation function is used to replace the Relu activation function,and the U-Net semantically segmented network is improved to advance the training efficiency and the accuracy of test set.The results show that the Kappa coefficient is improved from 0.658 to 0.886 and the classification accuracy of planting structure is improved to 93.7% when NDVI(Normalized Difference Vegetation Index)and AWEI(Automatic Water Extraction Index)are combined with the improved U-Net model algorithm.Combined with remote sensing and sensor measured data of soil moisture content,an inversion model of soil moisture content at different depths based on TVDI was established,which realized accurate inversion of soil moisture in the irrigated area.The inversion accuracy of each layer is above 94.25%,and the results are of high reliability.In view of the problems such as solidification of parameters,iteration prediction and less influencing factors in prediction of soil water content,using weather data acquired by web crawler and dynamic soil parameters returned by sensors to participate in the process of forecasting,a dynamic multifactor soil water content prediction model based on LSTM(long-term and short-term memory)was proposed.The results are less discretion,and the average of predictive accuracy rate is 98.7%,the root mean square error is as low as 1.447,and the coefficient of determination is 0.838.Based on the study above and the basic data of the optimized empirical formula for irrigation water requirement in the middle and lower reaches of the Yellow River,the image data of inversion and prediction are obtained.Based on ArcGIS secondary development platform and.NET framework,C# is chosen as the development language,three-tier architecture mode is used for component development,with WPF framework design interface,Fluent display Ribbon interface,which realized the automatic classification of planting structure,expression and prediction of soil moisture distribution in different depths,prediction of irrigation water requirement,statistics of irrigated area and some other functions.