The Land Use Classification Research Based On Multi-scale Segmentation

With the continuous development of high resolution remote sensing image,the technology of remote sensing observation has gradually matured.High spatial resolution is the general developing trend of remote sensing.In the low-moderate resolution remotely sensed images,the scale of the target is larger,and the details are blurred.While,in the high resolution images,the dimension of the ground object is small,the details are refined and it has clear relations with the surroundings,which will lay the good foundation for the processing and analysis of remotely sensed images.Therefore,the technology of high spatial resolution remote sensing has been applied in many fields,especially in the case of information extraction which has showed its important application value.The complex and varied information and complicated interference information in the high resolution remotely sensed image make the use of high resolution remote sensing image to extract the feature face enormous challenges.Therefore,the characteristics of the high resolution remotely sensed image should be understood in-depth,and the suitable information extraction technology should be explored,which can improve its application value.Based on this,the principle of multi-resolution segmentation were expounded,the scale parameter,spectral heterogeneity,shape heterogeneity and other factors were analyzed in this paper.The methods of edge detection were studied,and the edge information of the ground objects were integrated to participate in the multi-resolution segmentation to create the image object layers associated with practical objects.Based on the analysis of the characteristics of the objects,and the main features of the study area were extracted by combining threshold classification with fuzzy classification.In addition,the study area was conducted with the monitoring classification and unsupervised classification.Lastly,moderate amount of samples were chosen to calculate the confusion matrix to evaluate the accuracy of the results.Through the study,this paper obtained the following results:1)The results show that sobel operator can obtain higher quality edge of feature on the edge detection by comparing the results obtained by different algorithms.Aiming at the existing problem of automatically extracting edge,some special difficult edges were extracted by artificial method.This can provide edge data for multi-resolution segmentation.2)The image object layers were created by image segmentation based on further study the principle of multi-resolution segmentation technology and the important impact factors.The results show that the method can divided interference information and the adjacent pixels into the same homogeneous area,which can effectively reduce or eliminate noise interference,and solve the problem of local heterogeneity in the image.The problem that "the same object has different spectra","different objects have same spectra" can get better solved.Therefore,in the application of information extraction based on the high resolution remotely sensed image,the multi-resolution segmentation is undoubtedly a reliable new method.3)The edge information was integrated in the object image multi-resolution segmentation.On the basis of repeated experiments,the optimum scale of the ground object was determined by analyzing the relation between the biggest area of object and the scale.The results show that the quality of image segmentation has been greatly improved.It is quite clear that the edge is obvious between objects,the object is more effective and separable,and thus the accuracy of information extraction has been improved.4)Through the systematic analysis of the image object features and practical ground object features,extracting the main features of water,residents,vegetation,road and other object.The subliminal segmentation and fuzzy classification of the classification technique were applied to extract the six land objects in the study area,and the project graph of the six local objects in the study area was obtained.After completing the classification of the object oriented multi scale,the classified results were obtained by the monitoring classification and unsupervised classification.Finally,the evaluation of the precision analysis of three kinds of classification is carried out.Finally,select the evaluation results after computing confusion matrix of proper sample,which shows that all types of feature extraction with a high accuracy,the overall precision of 91%,the kappa coefficient of 87%.The results showed that ground object information extraction method with multi-scale segmentation has obvious advantages and good prospects in the high resolution remote sensing data applications.