Design And Test Of Kiwifruit Orchard Irrigation Decision Support System Based On UAV Multispectral

Water is an indispensable natural resource in agricultural production,and Soil Water Content(SWC)is a key factor in determining plant growth.Accurate estimation of SWC can provide a reliable reference for accurate irrigation.However,the unique planting conditions of kiwifruit orchards make it difficult to obtain better results with traditional SWC estimation methods in large fields,and irrigation strategies are not sufficiently detailed.In order to accurately estimate Root Soil Water Content(RSWC)of kiwifruit plants and develop precise irrigation strategies,this study discusses the feasibility of RSWC estimation and precise irrigation via an unmanned multispectral remote sensing platform combined with intelligent algorithms and wireless devices,using Xuxiang kiwifruit fruit trees with different irrigation treatments from 2021 to 2022.The study focuses on the acquisition of RSWC,multispectral,and meteorological data.Three RSWC inversion models were developed to achieve accurate estimation of RSWC and irrigation decision for single fruit trees.We designed and deployed the irrigation decision support system hardware and software to verify the accuracy of RSWC inversion model in practical application and the system execution effect,and realized the precise irrigation in the test area.The main work and conclusions of this paper are as follows:(1)Modeling vegetation indices screening.Using remote sensing and ground truth data,the changes of air humidity,soil moisture content and vegetation indices(VIs)during the experiment were counted.Pearson’s analysis and one-way ANOVA were performed on eight vegetation indices and RSWC to determine the sensitive vegetation indices.The results showed that the Renormalized Difference Vegetation Index(RDVI)and the green Normalized Difference Vegetation Index(g NDVI)showed good correlation(Pearson correlation coefficients of 0.744 and 0.525,respectively)and significance(P values of 0.007 and 0.015,respectively)with RSWC,and were suitable as inputs to the RSWC inversion model.(2)Inversion modeling.Three RSWC inversion models were established based on Multilayer Perceptron(MLP),Convolutional Regression Neural Network(CRNN)and Multiple Linear Regression,respectively,to investigate the effects of different sampling sizes on model accuracy,define two convolutional Convolution derived Vegetation Indices(CVIs)were defined and multiple regression models were built based on these indices.The results show that:the diagonal and area of the sampled area have high correlation with model accuracy(R2=0.991 and 0.993),and the optimal width of sampling is determined to be 300 px;the CVIs have high correlation with RSWC(Pearson correlation coefficients of-0.83 and 0.86,respectively).The coefficients of determination of the three RSWC inversion models on all samples were 0.565,0.743,and 0.636,with root mean square errors of 2.516%,0.887%,and1.219%,respectively.(3)Irrigation decision support system software and hardware construction.The irrigation system consists of 3 parts:data acquisition,decision and control,and execution and feedback.Its hardware includes UAV,weather station,computer,Lo Ra transceiver,solar controller,collector,JNBL 15 solenoid valve,YF-B5 sensor and micro sprinkler irrigation.The hardware system relies on Lo Ra wireless communication and signal lines to form a whole.The valve body receives computer commands and performs operations through the controller,and the computer judges the irrigation status through the information returned by the collector,ultimately forming a closed loop of control.Irrigation system software development including decision model implementation,control interface writing and thread assignment.The decision model uses meteorological data and standard crop take-off calculated by Penman-Monteith method as reference,estimates RSWC using MLP inverse model and interpolates the irrigation amount for different plants;The control interface is written in Py Qt5 and packaged as an executable file via pyinstaller.(4)Experimental verification of irrigation decision support system.Meteorological,soil and remote sensing data of the hardware deployment area were collected,and each data was automatically processed using the irrigation decision support system software to generate an irrigation strategy and executed by the hardware.The accuracy of RSWC estimated by the system and the change of soil water content after system execution were evaluated respectively.The results showed that the coefficient of determination of RSWC estimation by the irrigation system was 0.685,the root mean square error was 2.466%,and the absolute error was less than3.47%,which could estimate the spatial distribution of RSWC more accurately.After irrigation system decision and control irrigation,the original different RSWC(26%±8%)was controlled at an approximate level(34%±3%),and the variable irrigation target was reached.The amount of water used for variable irrigation in the test area was reduced from 4.12 m3 to 2.24 m~3 based on the Penman-Monteith method,a reduction of 46%.The irrigation control system can achieve variable irrigation for different RSWC targets and save water to some extent.