Cotton Moisture Condition Monitoring Based On Uav Thermal Infrared Canopy Temperature Characteristic Number

【Objectives】By monitoring crop water condition quickly and accurately,the optimal scheme of farmland water management can be made according to the monitoring results.It is of great significance to avoid or alleviate crop water stress,improve water use efficiency of farmland,and promote the development of precision agriculture.UAV thermal infrared remote sensing has the advantages of low cost,high flexibility and high resolution.It can acquire orthophotos of crop canopy temperature in real time and quickly.It is an important platform for large-scale monitoring of crop moisture conditions.The objective of this study is to construct a canopy temperature characteristic number of drip irrigated cotton crop based on UAV thermal infrared monitoring technology,which provides a method for nondestructive monitoring of drip irrigated cotton moisture status.【Methods】The experiment was carried out from 2021 to 2022.The cotton was treated with 80%（80%FC,I1）,70%（70%FC,I2）,60%（60%FC,I3,control group CK）and 50%（50%FC,I4）of the field water holding capacity of 0-40 cm soil profile before irrigation.Thermal infrared images of the cotton canopy were acquired using a drone equipped with a thermal infrared sensor the day before and the day after irrigation to extract canopy temperature information.The experiment also calculated characteristic numbers of crop canopy temperature and collected soil water content,stem flow and agronomic traits of cotton.The paper explores the effect of different algorithms for eliminating soil background from thermal infrared images on the accuracy of canopy temperature extraction.The paper analyzes the quantitative relationship between crop water stress index（CWSI）constructed using thermal infrared image temperature information and cotton water status.Finally,the paper tests the accuracy of the UAV thermal infrared-based cotton moisture monitoring model.【Results】（1）The paper explores an optimized algorithm for removing soil background and extracting canopy temperature from UAV thermal infrared images.The average accuracy of the two-year confusion matrix of the canopy images extracted by the three algorithms were as follows:the Otsu algorithm（72.01%,80.50%）<K-means algorithm（85.10%,89.50%）<SVM algorithm（92.24%,94.89%）.The temperature frequency histogram of the original thermal infrared image shows a bimodal distribution.And the canopy temperature frequency histogram after removing the soil background shows a single-peaked distribution.The two-year canopy temperature frequency histograms after removal of soil background by the SVM algorithm had the least dispersion,with minimum temperature standard deviations of 3.56°C and 3.43°C,respectively.Two-year correlation analysis showed that,compared with the Otsu and K-means algorithms,the SVM algorithm had the largest coefficient of determination R²（0.902,0.931）,the smallest root mean square error（1.344°C,1.452°C）,and the slope of the linear fit（1.017,1.009）closest to the 1:1 line.The results show that the SVM algorithm is the best algorithm for extracting the cotton canopy temperature after removing the soil background from the UAV thermal infrared images.（2）The paper compares the capability of CWSI based on thermal infrared image and traditional CWSI in cotton moisture monitoring.The highest daily growth of plant height and daily accumulation of aboveground dry matter were observed in the I1 treatment at the seedling and thunder stages.I1 treatment significantly increased by 14.97%to 40.78%and 11.31%to 37.43%over I3 treatment during the two reproductive periods.I2 treatment had the highest daily growth of stem thickness with a significant increase of 20.63%-28.71%over I3 treatment.The daily accumulation of aboveground dry matter and stem flow rate were the highest in I2 treatment at full flowering stage and full boll stage.I2 treatment significantly increased by 17.13%-65.26%and 11.28%-21.47%than I3 treatment during the two reproductive periods.Treatments I1 and I2 had a greater effect on soil moisture content in 0-40 cm at 33-75 days after emergence.Treatment I4 mainly affected soil moisture content in 0-30 cm at 40-75 days after emergence and had less effect on soil moisture content in 30-40 cm.Fitting results showed that there was significant or extremely significant negative correlation between the three CWSI and soil water content and stem flow rate.Among them,the highest coefficient of determination R²（0.702-0.709,0.497-0.509,0.403-0.773,0.327-0.810,0.901-0.917）was obtained for CWSI（CWSIsi）based on the simplified method of thermal infrared images.CWSIsi can better characterize the moisture status of cotton.When CWSIsi exceeded 0.436 and 0.453 at the seedling and shoot stages,respectively.cotton fields need to be irrigated when the soil moisture content of cotton is below 70%FC and68%FC at the seedling and bud stages,respectively.CWSIsi exceeded 0.506 and 0.500 at bloom and full bell stage,respectively,and irrigation was required when stem flow was below 20.210 cm-h^-1 and 20.384cm-h^-1,respectively.（3）Temperature feature number of UAV thermal infrared image based on various modeling methods to monitor cotton moisture.There was a significant or extremely significant correlation between the canopy temperature characteristic number of UAV thermal infrared image and soil moisture content and stem flow rate at seedling stage,bud stage,flowering stage and boll stage.The optimal response depth of canopy temperature characteristic number to soil moisture content was consistent with CWSIsi,which was 0-10 cm,0-20 cm,0-30 cm and 0-30 cm,respectively.The inversion of cotton moisture status by machine learning fusion of multiple canopy temperature signatures was best based on support vector regression（SVR）.（soil water content R²=0.891,RMSE=0.197%,MRE=1.636%,stem flow R²=0.920,RMSE0.858%,MRE=3.746%）,which was higher than the quadratic function model constructed by CWSIsi.Support vector regression（SVR）is the best method to retrieve cotton moisture status based on machine learning and fusing multiple canopy temperature characteristic numbers.The SVM algorithm was more accurate than the quadratic model of CWSIsi with soil water content R²=0.891,RMSE=0.197%,MRE=1.636%and stem flow R²=0.920,RMSE0.858%,MRE=3.746%.【Conclusions】（1）SVM algorithm is the optimal algorithm to eliminate the soil background of UAV thermal infrared image and extract crop canopy temperature.（2）I2 treatment can promote the growth and development of cotton and water absorption ablity.The CWSIsi constructed by using the canopy temperature information of the UAV thermal infrared image can better represent the moisture status of cotton.The threshold values of indicated irrigation in cotton seedling stage,bud stage,full flowering stage and full boll stage were 0.436,0.453,0.506 and 0.500,respectively.（3）A cotton moisture monitoring model based on the canopy temperature characteristic number of UAV thermal infrared image is constructed,and SVR achieves the highest accuracy.