Research On Extraction Method Of Distribution Information Of Peel Walnut Forest Based On Multi-source Remote Sensing

Walnut is the main plateau characteristic crop in Yunnan Province.Yunnan Province is also one of the birthplaces of walnut.The province is widely planted with walnuts.Among them,the walnut in Dali City is the most famous.Plateau characteristic agriculture is developed according to different geographical environments,and its development is directly related to the economic benefits and stability of the region.In order to achieve information monitoring of walnut forest,evaluation of planting environment suitability,and production evaluation,it is necessary to obtain accurate and timely information on spatial distribution.Because agricultural production activities generally have large areas,large regional differences,strong seasonality,and low economic benefits per unit area,the distribution of crops through traditional ground survey methods can no longer meet the actual needs.At present,the rapid development of remote sensing technology has brought effective solutions to this problem.Based on the comprehensive analysis of the phenological characteristics of walnut forest,this paper uses Sentinel data to construct multi-source data,and innovatively uses object-oriented extraction method for walnut forest information extraction research,and compares different multi-source data to construct walnut phenology.The degree of expression of the information and its impact on the extraction effect.The research results are as follows:(1)Research on information extraction of walnut forest based on time sernes vegetation index.In order to combine the high-precision extraction of walnut forest phenology with walnut phenological characteristics,the NDVI.RVI and DVI vegetation indices were constructed to construct time series,which can effectively reflect the phenological characteristics of walnut forest and make it easy to distinguish from other land types.Because the unique texture feature of walnut forest can be used as a large recognition feature,the object-oriented classification method is selected.After continuous optimization and optimization in the multi-scale segmentation process,the segmentation scale is set to 30 and the merge scale is set to 60.As the optimal segmentation scale.In order to express walnut forest information in the data source as much as possible,the texture information,DEM elevation information and vegetation index time series data are combined to construct multi-source data.Due to the data redundancy of the texture information,the first three bands representing its main information are retained as texture factors by PCA principal component analysis.Considering the original classification data as multi-source data,the main advantage of the support vector machine model is that it can obtain the main information from the multi-feature information,so the support vector machine model is used for extraction.The user accuracy of' the walnut forest was 84.8%.the drawing accuracy was 85.4%,the walnut forest error was 15.2%,and the leak error was 14.6%.The extraction based on time series vegetation index method is the best,because it can comprehensively reflect the phenological information of walnut forest,but there are data redundancy.complicated calculation,and low efficiency.(2)Study on the extraction method of walnut forest in the best phase.The key phenological period combined with walnut forest can extract its information quickly and conveniently based on a single phase.In order to find the best classification phase.according to the NDVI time series vegetation index values,the best time phase for walnut forest extraction was found in November by the maximum variance method.Combining DEM elevation factor,texture factor and vegetation index factor to construct multi-source data for object-oriented information extraction.Through the confusion matrix analysis,the overall classification accuracy is 83.54%,and the Kappa coefficient is 72.99%.The accuracy of walnut forest mapping is 90.51%,and the user accuracy is 75.61%.This method can quickly extract walnut forest,but its precision is not high.because the single phase data source can not fully express the phenological characteristics of walnut forest,making it easy to be confused with other land types.Research on the extraction method of walnut forest based on spring and autumn phases.Walnut forest is a broad-leaved deciduous forest.The original 26-band in the spring of May and autumn in November is used to carry out the three independent principal component information before the dimension reduction,and the basic data of multi-source classification is constructed by combining texture NDVI and DEM.The object-oriented walnut forest information extraction,the user accuracy of walnut forest extraction is 77.25%,the mapping accuracy is 94.2%,the walnut forest error is 22.75%,and the leakage error is 5.84%.Based on the spring and autumn phase extraction experiments,the extraction accuracy is between the time series vegetation index method and the best phase method,indicating that the multi-source data constructed can partially express the walnut forest information.(3)In general,all three methods can roughly extract walnut forest,but by comparing the extraction effects of three different phase multi-source data,the walnut planting area extracted by multiple time series vegetation index method is closer to the statistics.The total area of walnut forest in the yearbook indicates that the construction of source data that can fully express crop phenological information can effectively improve crop classification accuracy.