Research On Remote Sensing Monitoring Of Abandoned Farmland Based On Sentinel-2

Abandoned farmland has always been a serious problem all over the world.According to incomplete statistics,about 2.2 billion acres of farmland are abandoned,and 10.7% of the population is affected by hunger.In April 2020,affected by COVID-19,about 13 countries including Russia,Vietnam,Cambodia and Thailand announced restrictions on grain exports to ensure their own food supply.Therefore,Group of 20 convened an emergency meeting of agricultural ministers,calling on countries to maintain food exports and ensure the balance of world food supply and demand.The outbreak and rapid spread of COVID-19 have hindered international food trade,delayed domestic transportation,reduced the agricultural labor force,and posed a huge threat to food security in various countries.Based on this background,it is particularly important and urgent to carry out research on accurate monitoring of abandoned farmland,and to investigate the available farmland area and grain production capacity in China.This article takes Qi County as the research area and uses Sentinel-2 remote sensing image data in different phases from 2017 to 2019 to monitor changes in cultivated land and isolate abandoned farmland.First,the e Cognition software was used to extract the cultivated land of Qi County,including the segmentation and classification of the images.The basis for segmenting remote sensing images is to select segmentation parameters suitable for the area in order to optimize the segmentation results.Objectoriented image classification needs to be completed by constructing auxiliary samples of feature space in advance.Finally,the distribution range of cultivated land of Qi County was obtained.The verification of the accuracy of cultivated land extraction used pixel-based confusion matrix method.The results showed that the overall accuracy of this method for extracting cultivated land was 90.14%,the Kappa coefficient was 0.8026,and the accuracy of extracted cultivated land was ideal.On the basis of using the object-oriented classification method to extract the cultivated land,the joint change detection method was used to monitor the abandoned farmland.Through the NDVI change characteristics and threshold segmentation test in the study area,an appropriate threshold segmentation algorithm was selected for detection.In this paper,the extraction of abandoned farmland integrated two methods of intra-and inter-annual detection.That is,according to the changes of NDVI from spring to summer and summer to autumn,the abandoned farmland without planting and cultivated land with planting behavior were identified,and the judgement was based on the changes of surface coverage and spectral characteristics in different years.The combination of these two methods was the range of abandoned farmland.The extraction accuracy of abandoned farmland was verified to be 86.3%.The extraction results showed that from 2017 to 2019,the scale of abandoned farmland in Qi County had been shrinking,more fragmented,and more spatially dispersed.In general,abandoned farmland had been transformed from large-scale,complex-shaped plots into small-scale,regular-shaped,and scattered plots.Regarding its distance from the road and township centers,the analysis results show that the closer the roads and towns are,the higher the abandonment degree of cultivated land becomes.Such change trends indicate that the closer to roads,cities and towns,the lower the cost for farmer to go out and the more employment opportunities,which results in a larger amount of cultivated land being abandoned.