Research On Recognition Method Of Abandoned Land In Eastern Yunnan Based On Landsat Time Series Image

With the development of industrialization and urbanization,the phenomenon of cropland abandonment has become more and more prominent.Large-scale abandoned land leads to waste of cropland resources,which affects national food security.Therefore,it is important to identify abandoned land to ensure regional food security.With the rapid development of related technologies such as image processing,machine learning,and remote sensing big data,compared with traditional methods,it is more convenient to identify long-time and large-area of abandoned land by computer-related technologies.The Google Earth Engine cloud platform has massive remote sensing image information and fast cloud computing processing capability,which makes large-scale remote sensing data processing particularly efficient.Landsat remote sensing image has the advantages of long time series and high spatial resolution,which makes it the main data source of abandoned land identification.Therefore,taking eastern Yunnan as an example,this article relied on the Google Earth Engine cloud platform,based on the Landsat data from 2000 to 2020,used the random forest algorithm combined with the constructed stable sample set,generated the land use type coverage map,and used pixel-by-pixel time series analysis to identify abandoned land.The main contents and conclusions were as follows:（1）Stable sample set of land use types based on decision tree.The spectral index time series was used to judge the stable and changing sample points.The variance and coefficient of variation characteristics NDVI,RVI and BSI of remote sensing images were constructed,and the decision tree classifier was used to classify the remote sensing images in order to identify regions of stable and changing land use types.A certain number of sample points of different land use types were randomly generated based on the stable region and the existing land use product data,and the sample set generated by high-definition image inspection was used as the final data.The results showed that:The overall accuracy of the method for generating the stable region was 0.94,the precision was 0.97,the recall was 0.93,and F1 was 0.95,indicating that the method was feasible and achieved good precision effect.It could be used for sample collection in large-scale study area.The final stable sample set constructed in this paper included 600 cropland,80 grassland,600 woodland,70water bodies and 90 construction land,totaling 1440.（2）Remote sensing classification of land use based on random forest.Five types of features including spectrum,phenology,texture,terrain and tassel cap transformation were constructed.Based on the constructed feature set,the remote sensing images of different years in the study area were classified by using the random forest algorithm.The land use type coverage map of each year from 2000 to2020 was obtained,and the accuracy of the classification results was verified.The results showed that:The overall classification accuracy of each period was above0.90,and the Kappa coefficient was greater than 0.80,the classification effect was good,and it could be used for subsequent abandoned land identification.（3）Abandoned land identification based on pixel-by-pixel time series analysis.Firstly,the abandoned land identification rule was established,and the time series of classified images was constructed.Then,the abandoned land in each year was identified by the monitoring of land use change in adjacent years,and the abandoned land data set from 2001 to 2020 was generated.The first abandoned land year in 2020was calculated by the established abandoned land identification algorithm.Finally,the abandoned land identification results were tested and analyzed.Finally,the identification results of abandoned land were verified and the changes of the results were analyzed.This method can not only identify abandoned land year by year,but also calculate abandoned land starting time.The results showed that:By comparing the verification sample points obtained from the field investigation,the overall accuracy of this method in identifying abandoned land was 0.75,which had certain accuracy.From 2001 to 2010,the average area of abandoned land in the study area was 1140.97km~2,and from 2011 to 2020,the average area was 1692.44 km~2,which indicated that the degree of abandoned land in eastern Yunnan in recent 10 years was more serious than that in the previous 10 years.From 2014 to 2018,the abandoned land area and abandoned land rate in eastern Yunnan showed a downward trend as a whole.The abandoned land was concentrated in the west and southwest of Zhaotong.There was a large area of abandoned land in Qujing,and some areas had been abandoned for a long time.The abandoned land in the northern part of Wenshan was more serious in some central areas.