Research On Forest Land Change Detection Based On GF-2 Image In Zhangshanying Town,Yanqing District

GF-2 was the first domestically produced sub-meter high spatial resolution remote sensing data successfully launched in China's Gaofen series of satellites,which had wide application potential in land and resources monitoring.Forest resources are important ecological resources in China.Mastering their spatial distribution changes is of great significance in formulating plans for the development and utilization of forest resources and rationally organizing the development of forestry production.To promote the application of domestic Gaofen satellite in forest land change detection,based on the 2015and 2018 GF-2 images of Zhangshanying Town,Yanqing District,the three different change detection methods were used to obtain the land area change transfer matrix and distribution map of the study area in the second period:the image difference method based on pixel,the post-classification comparison method based on object-oriented classification,and the object-oriented change vector analysis(OCVA).Furthermore,through the accuracy evaluation and analysis of the change detection results,this research would provide tactical reference for the construction of forest land dynamic monitoring system based on GF-2 remote sensing images.The main conclusions of the research are as follows:(1)The red and blue bands were selected as difference band for the image difference method,the results showed that the difference map obtained by selecting the red band was more in line with the actual changes,while the change area with reduced gray value was not detected correctly when using the blue band.Finally the percentage change of the study area was calculated as 1.43%.The image difference method could basically determine the change rate and change area of the study area,and could provide a reference for quantitative change detection.(2)The post-classification comparison method based on object-oriented classification was used to detect changes in the study area.Object-oriented classification was performed on the two phase images,the shape factor and compactness factor were determined through experiments to be 0.2 and 0.6,then the ESP scale evaluation parameter tool was used to set the optimal scales for objece-oriented classification of GF-2 images were 635,550,450,and a rule set system was established on the segmentation scale to achieve classification.Finally,the results of the two-stage classification were combined to obtain the land change area transfer matrix and the land change distribution map of 2015-2018 Zhangshanying Town.The classification accuracy of GF-2 images in 2015 and 2018 were 82.96%and 84.63%,respectively.(3)Object-oriented change vector analysis(OCVA)was used to detect changes in the study area.Through the discussion of different segmentation modes and different constructions of change vector,etc.,changes were detected in a typical study area to determine the optimal OCVA.The results showed that:(1)The contour of the object segmented by the multi-temporal combination segmentation mode was closer to the actual feature boundary,and the detection accuracy was 92.01%higher than 88.95%of the multi-temporal split segmentation mode;(2)The OCVA method based on the weighted combination Euclidean Distance could reduce some false changes caused by the GF-2 image itself and improve the accuracy(92.63%);(3)The optimal OCVA method was a method of calculating change vector using weighted combination in multi-temporal combination segmentation mode.(4)The optimal OCVA method,the OCVA method combined with object-oriented classification were used in the entire study area for the change detection research,and compared with the post-classification comparison method based on object-oriented classification and image difference method.(1)From the change result map,there were many false changes between shrub and forest land in the change detection results after object-oriented classification,therefore,the OCVA method could effectively reduce the false caused by the lower classification accuracy in the object-oriented classification method;(2)The change detection accuracy of image difference method,object-oriented classification method,OCVA method,and OCVA method combined with object-oriented classification were 87.13%,88.50%,90.88%and92.50%,respectively.Therefore,the OCVA method combined with object-oriented classification had the best detection accuracy and effect.(5)According to the change detection results obtained by the OCVA method combined with object-oriented classification,the area changes of forest land and non-forest land in the past three years were calculated.It was found that the area of forest land decreased by 200.04 hm~2,among which forest land decreased the most,while other forest land increased slightly.These changes were closely related to the construction of venues,urban-rural integration,and afforestation.