Potato Growth Monitoring And Evaluation Based On UAV Remote Sensing

Timely and accurate acquisition of crop growth conditions and environmental information in the field is a prerequisite and basis for accurate crop management.As a flexible and efficient technology for acquiring environmental information and crop growth information on farmland,UAV remote sensing has been widely used in agricultural production and scientific research in recent years.With the advent of the Agriculture 4.0 era,UAV remote sensing has become an important part of intelligent agriculture,providing data support and decision-making basis for intelligent agricultural management.The field trial of this study was conducted in Hale Town,Wuchuan County,Hohhot City,Inner Mongolia Autonomous Region,2022.Potatoes grown in different nitrogen gradients were used as the research object,and the DJI Genie 4 RTK UAV with visible light sensors was used to acquire RGB images of potato seedling,tuber formation,tuber expansion and starch accumulation stages,and agronomic parameters such as plant height,leaf area index and above-ground biomass were collected simultaneously on the ground to construct a one-dimensional linear regression model,partial least squares regression and random forest regression models,respectively,based on The models for estimating cover,plant height,leaf area index and above-ground biomass of potatoes at key fertility stages were constructed based on one-dimensional linear regression models,partial least squares regression and random forest regression models,respectively,and the optimal estimation models were selected through comparative analysis of the models.The main conclusions are as follows:1.In order to achieve dynamic monitoring of potatoes throughout their reproductive life,the ratio of green pixels to all pixel points was calculated using the CIVE colour vegetation index as a segmentation criterion at the seedling stage,combined with the Otsu threshold segmentation method to remove the soil background,to obtain the potato cover,and the results showed that the highest potato cover（56.16%）and its uniformity was better in the soil-N balance-based treatment（30.69）.2.Potato plant height(H_dsm)was extracted by digital surface model（DSM）from UAV-visible images at tuber formation,tuber expansion,starch accumulation and full fertility stages,and estimated by single and full fertility stages,showing that the R² fit between estimated plant height(H_dsm)and measured plant height（H）was above 0.74 and RMSE was less than 5.8 The best validation was achieved at full fertility,with a coefficient of determination R² of 0.82 and an RMSE of 4.63 cm.3.A partial least squares regression model and a random forest regression model for potato were constructed from vegetation indices,texture features,and spectral indices fused with texture features,respectively,with the measured values of LAI and AGB.The results showed that the growth indicators（LAI,AGB）were in good agreement with the calculated spectral indices and extracted texture features,and that the machine learning input parameters influenced the accuracy of potato growth indicator monitori ng.For the three selected input variables,the model predictions were more accurate using spectral indices fused with texture features than using spectral indices or texture features alone.The best performing algorithm was the random forest algorithm for a single fertility period as well as for the entire fertility period.The best LAI observation period was the starch accumulation period,with a coefficient of determination R² of 0.81 and RMSE of0.206,and the best AGB observation period was the full fertility period,with a coefficient of determination R² of 0.94 and RMSE of 340.89.4.Analysing the effect of each indicator on yield,it was found that different fertiliser application methods delayed potato fertility and thus affected yield.This was evident in the fact that poorly grown potatoes had greater H,AGB and LAI during the starch accumulation period,so that the indicators did not show a monotonic trend in relation to yield,but were in dynamic change.The above results show that this study used UAVs with visible cameras to acquire remote sensing images,analysed the agronomic parameters of potatoes during key fertility periods,constructed quantitative prediction models,realised the dynamic monitoring of potatoes throughout the fertility period,and finally analysed the impact of each index on yield.It can provide a reference for accurate monitoring of potato growth parameters using remote sensing technology.