Research On Cotton Insect Pest Monitoring Based On UAV Multi-spectral Remote Sensing Image

As the cotton production and planting area with the largest proportion and the highest yield in China,Xinjiang area accounts for 76.09% of the country’s planting areas.Ensuring the steady growth,yield growth and quality improvement of crops has always been a topic of hot scientific and technological research in recent years.The frequent occurrence and spread of pests and diseases have brought non-negligible losses to cotton growers.With the help of UAV remote sensing platform,remote sensing image data with high temporal and spatial resolution can be obtained efficiently and quickly,so that the growth of cotton can be predicted,and the pest area can be monitored in real time,which is of great significance for the rapid response to the occurrence of pests in cotton fields and precise cotton field management.Based on this,this paper uses the significant advantages of UAV remote sensing technology platform in timeliness and resolution to dynamically monitor the occurrence of cotton pests,and fully combines the spectral characteristics of the cotton field affected by pests and environmental factors.In-depth research has been carried out from three perspectives,the identification of spectral sensitive bands,the analysis of influencing factors,and the dynamic monitoring of the occurrence area of pests.The research on the dynamic monitoring and identification technology of cotton pests has been carried out.The main research objectives include the following aspects:(1)Insect pest spectrum sensitive band identificationUsing an advanced drone multi-channel spectral data remote sensing platform,the vegetation of the regional cotton field is used as the main spectral data research and monitoring object,and the data of various cotton field pests and vegetation spectral regression indexes at different time nodes and different environmental factors are integrated.Analyze and compare,analyze different insect pest spectral regression indexes,and use Logistic spectral regression analysis method to obtain NDVI(Normalized Cotton Field Vegetation Index),EVI(Enhanced Cotton Field Vegetation Index)and SAVI(Comprehensive Soil Regulation Cotton Field)according to the screening results.(Vegetation Index)A total of 3 remote sensing factors to estimate the severity of cotton pests are used to construct a pest diagnosis model based on the vegetation index and the sensitive band of the leaf spectrum.The cotton aphids,cotton spider mite,and cotton bollworm identification models jointly established by the SAVI(Comprehensive Soil Regulation Cotton Field Vegetation Index)model and the NDVI(Normalized Cotton Field Vegetation Index)model are the optimal models.The accuracy of training samples and test samples are accurate.The rates reached 93.7% and90.5% respectively,and the recall rate and F1 value were 96.6% and 93.5%;the Logistic model achieved a high level of accuracy in monitoring the pest areas of cotton fields,and accurately identified the occurrence areas of cotton aphids,cotton red spiders,and cotton bollworms.The degrees reached 94.2%,85.1%,and 66.3% respectively.From the perspective of the optimal model monitoring results,the cotton field pest identification model based on the Logistic model in this study can realize the monitoring and identification of cotton field pests and provide decision-making for precision plant protection and farmland management.(2)Impact factor analysis1)There are still limiting factors in using UAV multi-spectral remote sensing to monitor pests.For example,there are differences in the spectral data obtained at different times and under different weather conditions.How to eliminate the differences is a major and difficult problem in the field of multi-spectral remote sensing to monitor pests.2)Climatic and meteorological conditions,cotton field management methods,and plant protection and protection conditions are related to the occurrence of cotton field pests and become key factors in the dynamic changes of pests.The image data uses BP to predict the accuracy of the radiation correction error.For the wavelength data in areas where the frequency fluctuates greatly,the relative error of the test set is controlled at 10.5%.The model built by the BP method makes the reflectance correction value and the measured error evenly distributed,Has a good model fitting accuracy.The correlation coefficients reached 0.945,0.906,0.913,0.875,respectively;The results show that the regression model based on meteorological data can better reflect the correlation between the occurrence of insect pests and climatic and meteorological factors,and can provide decision-making data for precision plant protection operations.(3)Dynamic monitoring of pest areasThe dynamic monitoring method of pest multi-temporal remote sensing technology is studied.Based on the multi-phase image data collected and acquired by automated drones,the pest identification model analyzed in this paper is used to analyze and extract the pest image information of each phase.On this basis,autocorrelation time series detection(CCF)is used to the dynamic area of pest occurrence was identified,and the exponential curve detection model of insect pests changing with time was established to realize the dynamic monitoring of the area of pest occurrence.From the correlation coefficient,the standard error value and the dynamic change characteristics of the three insect pests,the occurrence and spread of the cotton aphids,the cotton red spider and the cotton bollworm all have similar conditions.The research of this subject mainly realizes the rapid identification and dynamic monitoring of cotton field pests within a certain area.The research results can provide methods and references for the fine monitoring of cotton field pests and the dynamic trend prediction analysis,and provide for plant protection and preventive work in advance.