Research On Remote Sensing Monitoring And Analysis Of Tobacco Region In Lianghe County Based On Sentinel-2 Data

Tobacco is a traditional agricultural cash crop and a highland specialty crop in Yunnan Province,and the tobacco industry is the primary pillar industry in Yunnan Province,whose development directly affects the industrial development,economic benefits and stability of Yunnan Province.The tobacco industry in Yunnan Province is currently facing digital high-quality transformation,and tobacco agriculture,as the cornerstone of the tobacco industry chain,currently has problems such as tobacco planting area and spatial information are difficult to effectively monitor and manage,and traditional ground survey methods cannot meet realistic needs.The application of remote sensing technology in crop classification identification and area monitoring research provides an effective solution to the problems faced by tobacco agriculture in Yunnan Province.In this study,Lianghe County,a representative high-quality tobacco planting area,was selected as the study area,and the advantages of Sentinel-2 satellite image data in high temporal and spatial resolution were used to combine GF-2 remote sensing images and DEM data to construct a time-series remote sensing image dataset covering the complete growing period of tobacco in the study area,from which time-series vegetation indices,red-edge indices and spectral feature curves were extracted to synthesize After that,we compare and analyze the extraction effect of tobacco planting area by CNN,RF and SVM methods and red-edge band,and finally apply RF method to extract tobacco planting area and spatial distribution information in Lianghe County from 2020 to 2021,so as to provide theoretical,practical,and practical information for the continuous dynamic monitoring of tobacco planting.It provides theoretical and methodological basis as well as practical significance for continuous dynamic monitoring of tobacco planting.The main research results are as follows:(1)Construction of tobacco classification feature parameter set in the study area.Constructing a set of tobacco classification parameters in the study area by applying multi-period Sentinel-2 satellite remote sensing data to extract vegetation indices and spectral characteristics of the study area,analyzing tobacco weather characteristics,combining 28 parameters such as texture characteristics and terrain characteristics extracted from GF-2 data,and evaluating and selecting each parameter combination by applying J-M distance,and finally proposing a combination of spectral,index,weather,and terrain parameters.A control group was set up to analyze the effect of red-edge band/index on tobacco identification and extraction.The analysis found that the difference between tobacco and other features,especially other crops,in April was the best window for information extraction;subsequent classification results showed that the red-edge band/index was sensitive to vegetation information and could better reflect tobacco identification information,effectively improving the accuracy of tobacco classification and planting information extraction,and theandcoefficients of extraction results improved by 3.27%and 0.0223.(2)Optimal classification method selection.In this study,three classification methods,CNN,RF and SVM,were applied to classify and extract the features in the study area.Typical areas with sufficient representativeness were selected,and the classification accuracy and effect of the three methods were analyzed by visual interpretation and confusion matrix in a comprehensive comparison.The results showed that all three methods could better extract the tobacco planting area,and theof SVM,RF and CNN classification results were 91.68%,93.60%and 95.80%,respectively,withcoefficients of 0.8858,0.9237 and 0.9319,and the time consumed was 27 min,30 min and 6h23 min,respectively.After comprehensive analysis,it was concluded that RF classification method better balanced accuracy and efficiency,and finally applied RF classification method to extract tobacco planting area and spatial distribution information in the study area in 2020-2021,and the accuracy was evaluated by using the confusion matrix,and the total planting area of 3004.56 hm~2was extracted,with an error of 1.51%compared with the actual area,of 93.30%,andcoefficient of0.9174,andwere both high,and the extraction effect was considered to be more satisfactory.(3)Spatial distribution of tobacco.Spatial analysis of the extracted tobacco area combined with the DEM of the study area and the actual statistics of each township showed that the extracted results of this study were consistent with the actual distribution of tobacco cultivation in the study area,and that tobacco in the study area was mainly cultivated in human habitable areas such as dams and foothills with an altitude of 812-1200 m,a slope of≤15°,and abundant human resources,with reasonable distribution,which is conducive to the production and management of tobacco and can obtain better It is reasonably distributed and conducive to tobacco production and management,and can yield good economic benefits.This study proposes a proven tobacco remote sensing classification identification and planting information extraction method for tobacco planting supervision in the study area,which provides a theoretical and methodological basis for real-time,accurate and continuous dynamic monitoring of tobacco planting in the study area,and at the same time provides basic data and decision support for relevant departments to understand tobacco planting,production and management,and to a certain extent provides a priori knowledge for realizing lean tobacco production and promoting digital transformation of tobacco agriculture,with certain theoretical and practical significance.