Study On Detection Method Of Forest Cover Change Based On GF-1 WFV Data

Forest is the largest ecosystem on the landecosystem and plays an importantrolein air purification,regulating climate,water conservation,reducing winddamage.The change of forest in quantity or quality has important effect on the change of ecologicalenvironment,biological diversity and global climate.Timelyand accuracy acquisition of forest change information is ofgreat significance tothe study on the environmental change and forest managementplanning.In this paper,the GF-1 satellite 16 m resolution multi-spectral wide(WFV)imageshave been selected as the data source for forest cover change identification.In order to explore the GF-1 WFV data on forestcover change detection method in China,two forest cover change detection methods have been constructed after the selected GF-1 WFV images had been processed by radiationcorrection,orthorectification,image registration and relative radiation normalization.One is the kernel principalcomponent analysis(KPCA)change detection method.Firstly,the datawas processed by the KPCA method,the difference image was constructed by the first principal component,and then the OTSU method has been selected to automaticallycalculate the threshold and extract the information of forestcover change in the study area.The other method is based on the time series of Integrated Forest Z-Score(IFZ)statistical characteristics.Firstly,to obtain accurateforest pixels,the forest training samples were extracted by using thedarkest target forest peak TDA(Trainning Data Automation)method after the darkest targets pixels including the water and shadow have been masked.Secondly,the IFZ index has been constructed using the selected forest training samples.At the end,the sequentially VCT(Vegetation Change Tracker)method has been used to extract the forest cover change information by tracking the pixels in different satellite images.The main results and conclusions of this paper are as follows:(1)Based on the characteristics of GF-1 WFV multispectral images,the HOT(Haze Optimized Transformation)method can be used to identify and eliminate the pixels covering by cloud and shadow effectively.(2)The GF-1 WFV imageshave been normalized by using the relative weighted radiation detection method(IR-MAD).The mean reflectance of the four bands of the target image was similar to the meanvalue of the relative band'sof the reference image.It shows that the radiation difference between different time's GF-1 WFV images can be eliminated by the method.(3)Based on the kernel principal component analysis(KPCA),the Gaussian kernel function was used to extract the feature of the normalized GF-1 WFV images.The difference between the first principal component was calculated by OSTU method.The experiment shows that the first principal component after KPCA processing contains the maximum amount of effective information,which is more than 90%.The method reduces the noise and makes the difference change information more concentrated,which is helpful to improve the change monitoring precision.The overall accuracy of the three study areas was above 86% respectively.(4)The forest training samples were extracted based on the TDA(Trainning Data Automation)method.Not only the forest peak was developed,but also the forest training samples can be identified in the study areaswhen the window size was set to 400 × 400 pixels after removing the dark objects such as water bodies and shadows.And then the annual IFZ index was constructed,and the sequentially VCT(Vegetation Change Tracker)method has been used to extract the forest cover change area.The results show that the overall accuracy of the three study areas is higher than that of the KPCA method by 3.05%,0.54% and 0.74% respectively.Which shows the IFZ index change detection method is better than that of the KPCA change detection method.