Forest Change Detection Based On Direction-first Change Vector Analysis

Forests are an important guarantee of global sustainable development.Accurate acquisition of changes in forest is very important for the study of biodiversity,ecological environment,and climate change.Remotely sensed images with low and medium spatial resolution have been widely used in the forest change detection at regional and global scales.However,due to the limitation of spatial resolution,the accuracy of detecting small forest changed area is poor.Using remotely sensed images with high spatial resolution can detect forest changes more accurately,but such images usually only have blue,green,red,and near-infrared bands.The forest changes are sensitive to limited bands,making it impossible to obtain complete forest change types.The red-edge band is a sensitive band of forest changes.Using high-resolution remotely sensed images with the red-edge band can detect more forest change types while ensuring the accuracy of detecting small forest changed area.Considering the influence of the spatial resolution and band ranges on forest change detection,a method called direction-first change vector analysis(DFCVA)is proposed.Sentinel-2 satellite images in the typical area of Huangshan was used as the main data source,and compared with the classic change vector analysis(CVA)method to analyze the applicability of DFCVA in the forest change detection based on highresolution remotely sensed images.The main research contents and conclusions of the study are as follows:(1)Analysis of the characteristics of forest change types.The causes of forest change are analyzed,and the forest change types within the range of optical remotely sensed image detection capability in the study area were clarified.Based on the types involved,relevant spectral samples are selected on remotely sensed images.After analyzing the obtained spectral response curve,the sensitive band to forest change is determined,which is convenient for the subsequent calculation of the change vector.(2)Construction of DFCVA forest change detection method.Classical CVA uses the magnitude of the change vector(that is,the modulus of the vector)to determine the change pixel first,and then uses the direction of the change vector to determine the change type.In order to weaken the influence of the unchanged area,DFCVA preferentially uses the direction information to obtain the change type,and then determines the final changed pixels by the magnitude.This method is composed of three parts: obtaining candidate changed pixels,giving the forest change types of candidate changed pixels,and determining the final changed pixels for each forest change type.Among them,the first part obtains candidate changed pixels from all pixels through the global magnitude threshold;the second part simulates the change vector information of each change type,constructs a look-up table of the forest change types,and then each direction range obtained by clustering the direction of candidate change pixels is given the forest change type;the third part determines the final changed pixels of each change type based on the local magnitude threshold at each direction range,and then uses the edge contour detection method to post-process the results,which can further improve accuracy.(3)Sensitivity analysis of key parameters of DFCVA.The key parameters of DFCVA in three parts are the global magnitude threshold,the number of clusters and the local magnitude threshold.The relationship between the global magnitude threshold T and the accuracy of results is analyzed to provide support for using automatically determined T to obtain candidate changed pixels.The influence of the number of clusters k on the accuracy of results is analyzed to provide a basis for choosing an appropriate k to determine the direction ranges.The relationship between the local magnitude threshold and the accuracy of results is analyzed,and it is shown that the automatically determined local magnitude threshold can be used to obtain a highly accurate change detection result.The comparison of the change detection results before and after the post-processing shows that the post-processing can effectively remove meaningless small change area,thereby improving the accuracy of forest change detection.(4)Comparison between DFCVA and the classic CVA.The accuracy of DFCVA and the classic CVA is compared in terms of automatic magnitude threshold and optimal magnitude threshold.The difference in accuracy of the automatic threshold reflects the difference in accuracy between the two in practical applications.Compared with DFCVA,due to the limitation of only one magnitude threshold,CVA has not been able to detect all change types on the image.The difference in the accuracy of the optimal threshold reflects the difference in the upper limit of the theoretical accuracy of the two.Overall,the accuracy of DFCVA is higher than that of CVA.Compared with CVA,DFCVA has advantages in the forest changes detection.This main contribution of the study is the construction of the DFCVA method to detect forest changes on high-resolution remotely sensed images.Through the sensitivity analysis of the key parameters of DFCVA,the relationship between the magnitude of the change vector ρ,the direction of the change vector θ and the forest change types is clarified.By simulating the change vector information of different forest change types,a look-up table was constructed.Compared with CVA,DFCVA first determined the change type according to θ,and then set different magnitude threshold for each change type.DFCVA can significantly improve the accuracy of change detection results in the determination of changed pixels and change types and have certain theoretical innovation and practical application value.