Individual Tree Locating Method Based On CV Model From High Resolution Satellite Imagery

Individual tree positioning is to obtain individual tree position coordinates to create a individual tree database,which is a basic and meaningful work for realizing precise forestry,especially the intensive management of urban trees.However,due to the wide spatial distribution,long growth cycle and huge number of individual trees,it is difficult to complete a large number of tree-level forest surveys(including individual tree location)by traditional ground-based manual surveys.For this reason,since the mid-1980 s,scholars have proposed and carried out research based on remote sensing images for individual tree location.Since high-resolution satellite images have the advantages of large size,easy accessibility and economy,research on individual tree positioning methods based on high-resolution satellite images has been a hot topic of interest.Due to the complexity of the terrain and the different sizes,close arrangements and interlocking canopies of trees,the remote sensing image interpretation may cause over-segmentation or under-segmentation within individual canopy and between overlapping canopies,resulting in commission and omission of individual tree positioning.The research shows that the accuracy of the current remote sensing image-based individual tree location method is difficult to meet the practical requirements,and there is no method that can be adapted to various forest stands,and new,more effective and stable individual tree location methods need to be researched.To this end,based on the summary of the process and preliminary work of individual tree localization based on remote sensing images,this paper attempts to apply the image segmentation technique based on CV model to individual tree localization,drawing on the successful experience of application in other fields.Firstly,the greenness segmentation and Gaussian filtering are used to extract forest area and complete the pre-processing required for individual tree localization;secondly,the initial contour is automatically obtained by an algorithm combining the morphology of the tree canopy and its image spectral features;then,the CV model is used to construct the level set function and iterate the initial contour line to obtain the individual tree canopy contour;finally,the individual tree location information is calculated.In order to test the effectiveness of CV model individual tree positioning method,a corresponding experimental platform was developed and seven high-resolution satellite images of different types(coniferous forest,broad-leaved forest,economic forest and non-forest stand)were selected to conduct a comparative analysis with the traditional individual tree positioning method.The results show that the CV model individual tree location method has a higher matching rate than the traditional methods such as gradient watershed method,marker watershed method and local maximum method,with an average increase of nearly 23%.This paper proposes and implements the CV model individual tree positioning method with automatic initial profile setting,which can better handle the connection and overlap of tree crowns and has better positioning effect,showing good potential for application.