Research On Remote Sensing Identification Method Of Cultivated Land And Typical Crops In The Central Plains,China

Agriculture is the basis of human existence and the primary condition for all production.As a large country with a population of more than 1.4 billion people,it is China’s basic national policy to fully guarantee the safety of food production.As an important part of agricultural production,information on the spatial distribution of cultivated land and crops is one of the important data sources for agricultural production monitoring,crop yield estimation,agricultural policy formulation and socio-economic planning.In order to ensure food security,the state has introduced a series of policies and regulations aimed at safeguarding agricultural production,such as basic cultivated land protection and agricultural insurance,which have put forward higher requirements for large-area,rapid and accurate access to information on the spatial distribution of cultivated land and crops.The traditional large-scale statistical survey on cultivated land and crop cultivation area is mainly conducted by combining visual interpretation of remote sensing images and field mapping.Although the data obtained by this method are highly accurate,there are shortcomings such as large internal and external workload,time-consuming and laborious.Therefore,this paper takes remote sensing precision recognition of crops and cultivated land as the starting point,selects the experimental research area in Henan Province,a major grain-producing province in China,carries out the research on remote sensing automatic recognition of typical crops and cultivated land based on multispectral remote sensing data,and conducts in-depth research on the missing time sequence in short weathering period,intra-crop differences and cross-site differences,as well as crop recognition and precise extraction of cultivated land under the influence of mixed image elements,so as to provide theoretical basis and data support for agricultural insurance,basic cultivated land protection and other policies and regulations.The main research and findings of this paper are as follows:(1)An improved artificial immune network crop recognition algorithm based on dispersed vegetation index genetic chain was constructedIn the process of remote sensing identification of typical crops in the Central Plains region,such as winter wheat and peanut,the influence of clouds and fog and other climatic effects can easily lead to the lack of data on the timing of key crop phenological periods,while the intra-crop differences caused by changes in crop species,planting environment and other factors can affect the accuracy of crop identification.To address the above problems,this paper introduces the Marxian distance to improve the similarity measure,and uses Sentinel-2multispectral remote sensing data as the data source to construct an improved artificial immune network crop recognition algorithm based on dispersed vegetation index genetic chain for remote sensing extraction of spatial distribution of winter wheat and summer peanut in Zhengyang County and winter wheat in Huai Bin County.Combined with the validation of the measured data,we found that the overall accuracy of crop recognition accuracy of the algorithm in this paper is better than 96%,and the Kappa coefficient is better than 0.91.Especially in the case of missing key timing data of crops,the accuracy is improved by up to22.79% compared with the random forest algorithm and support vector machine algorithm,and it can effectively overcome the effects of missing timing data,intra-class differences of crops and mixed image elements,and has stronger Robustness.Based on the crop identification algorithm constructed in this paper,the winter wheat in Zhengyang County with a complete timing sequence of phenological periods was selected as the research object,and the identification research was conducted under the situations of missing remote sensing data of single month,missing remote sensing data of multiple months and missing remote sensing data of different phenological periods,etc.The results show that the method in this paper can still maintain a good crop identification accuracy(overall accuracy≥97% under missing timing sequence,Kappa coefficient≥0.79).(2)A cultivated land extraction method considering time-series vector characteristics of crop growth cycle was constructedIn response to the problem that the complex cropping structure and diverse maturation systems of cultivated land in the Central Plains,as well as the mixed artificial woodlands,abandoned grasslands,and artificial buildings(structures)in and around the cultivated land,cause the confusion of spectra between different land types.This paper proposes to use the time series of cropland vegetation index and spectral features as the vector parameters for solving the five features of cosine(Cos),distance(Dis),maximum(Max),minimum(Min)and extreme difference(Ran),and constructs a cultivated land extraction method considering time-series vector characteristics of crop growth cycle,which integrates the recognition features of cropland under complex cropping structure and effectively expands the features between land classes.The difference term is effectively expanded.The results show that the overall accuracy of this method reaches up to96.62% and the Kappa coefficient reaches up to 0.95.Compared with the artificial neural network algorithm,maximum likelihood estimation algorithm and support vector machine algorithm,the overall accuracy is improved up to 25.43%,and it can also effectively suppress the noise points.It can also effectively suppress the generation of noise points,strip the non-cultivated factors such as grassland and woodland to a greater extent,weaken the influence of mixed image elements,and have high robustness to different ground covers.To verify the accuracy of this paper’s cultivated land extraction model at different spatial and temporal scales,the scale of the study area was extended to the South Taihang region,and the results show that the overall accuracy of this paper’s method is≥84%,with a Kappa coefficient≥0.81,and the cultivated land extraction calculation can be performed for multiple years using a single year of pre-training data.The statistical analysis of the spatial and temporal changes of cultivated land and landscape pattern changes in the South Taihang region from2016 to 2020 using the method of this paper,and the produced annual cultivated land distribution maps can meet the practical application needs of relevant government management departments.The innovative points of this paper are as follows.(1)An improved artificial immune network crop recognition algorithm based on dispersed vegetation index genetic chain was established.By introducing dispersion index genetic chain and improving the similarity measure with reference to the Marxian distance,it can effectively weaken the effects of missing time-series data and intra-crop differences,and improve the accuracy and robustness of typical crop recognition.(2)A cultivated land extraction method considering time-series vector characteristics of crop growth cycle was constructed.By introducing the vector concept to integrate the features of cultivated land with complex planting structure,the small differences of spectral features between land types are expanded,and the mutual interference of mixed image elements is eliminated to improve the accuracy and robustness of cultivated land extraction,and it has the advantages of less training samples and easier calculation.(3)The contribution of vegetation indices in remote sensing identification of crops and cultivated land was verified.For the process of constructing genetic chains and temporal vector features of decentralized vegetation indices,the weights of temporal features of vegetation indices are analyzed and verified by examples,and the contributions of vegetation indices under different temporal nodes are summarized to provide references for subsequent studies.