Object-oriented Classification Of Tropical Forest Based On Multi-level Segmentation

Forest resources monitoring is the important content of forest resources conservation,and forest classification is an important job of forest resource monitoring based on remote sensing technology.Low resolution remote sensing image is unable to provide accurate information such as the feature space,texture,shape,so the accuracy of forest resources monitoring is restricted.In recent years,with the continuous development of remote sensing technology,the spatial resolution of remote sensing image is becoming higher and higher,high resolution remote sensing image not only provides abundant spectral information but also contains rich texture,space and shape images feature information.Therefore,high resolution remote sensing image provides more help to the implementation of the forest resources monitoring.But the traditional classification method mainly use spectral information in remote sensing image classification,and cannot make full use of the rich information which high resolution remote sensing image provided,even may cause data redundancy and the phenomenon of "salt and pepper effect" is more serious,result of low accurately of forest resources information extraction and classification.Aiming at these problems,object-oriented classification method is proposed.The method can not only use spectral information,but also use its texture,space and shape information,which extracted from the high-resolution remote sensing image,based on segmentation images.At the same time,the method can give full play to the characteristics of high resolution remote sensing image,getting all kinds of feature information needed quickly,accurately,and that the method can avoid the phenomenon of "salt and pepper",thus improving the accuracy of feature classification.Therefore,the object-oriented method is gradually got attention and application in forest resources monitoring field.The object-oriented multi-scale classification method was used in this paper to information extraction of tropical forest in Hainan Bawangling National Reserve,based on SPOT6 high resolution remote sensing image.And this paper mainly discusses the acquisition method of segmentation scale and parameters as well as the establish process of multi-scale classification rules.The segmentation results showed that: in order to obtain optimal segmentation parameters for different forest land covers,the segmentations achieved good results by using the ESP(Estimating the Scale Parameter)tool,which is timesaving and could offer evidence for obtaining optimal segmentation parameters under different levels as well as avoid the effect of subjective factor.When it comes to feature selection,the study acquired classification feature rule set by using expert knowledge and statistical methods,according to the analysis and extraction of texture and shape features of the image in the study area.The classification results showed that the object oriented classification based on multi-scale achieved total accuracy of 84.46%,which was higher than object-based single-scale nearest classification with 75.06%.Therefore,the object-oriented multi-scale classification method is suitable for tropical forest vegetation classification research,whose structure is complex and has a wide variety of vegetables,making up for the less of the extraction researches of tropical forest resource information based on object-oriented method,as well as providing reliable evidence for tropical forest resources monitoring research.