Study On Extraction Of Rapideye Image Of Eucalyptus Forest Information Based On Object Oriented Technology

Eucalyptus is one of the most important tree species in China, has the characteristics of fast growth, the most widely used, and have the advantages of economic benefit, social benefit and ecological benefit. In the new situation of Guangxi forestry pulp paper integration of industrial development, the emergence of intensive planting Eucalyptus trees, the planting scale, scope and rapidly expanding area. On the other hand, because the eucalyptus trees, cutting and promoting alternative too frequently, and will have a certain ecological threats, such as through the traditional method of sampling investigation and the aggregate level is difficult to achieve effective monitoring of Eucalyptus planting. Therefore, the remote sensing technology, using object oriented image analysis method, timely and accurate grasp of Eucalyptus planting production, including planting Eucalyptus quantity, quality and dynamics of information, which to improve Eucalyptus planting, production, operation and management have great help, and to promote the cultivation of Eucalyptus, ecological environment protection and economic and social sustainable development of great significance.The peak of Guangxi state owned forest farm as the research area, using RapiEye satellite remote sensing image based information source, combined with the results of GPS field survey data, the eucalyptus forest remote sensing information extraction for target, ERDAS, ArcGIS, ENVI, eCognition platform based on the classification method, object-oriented image analysis techniques and knowledge rules created by the combination of extraction the study area of Eucalyptus planting information. Specific research ideas are: first, using the object-oriented multi-scale image segmentation technique for RapidEye image segmentation, construction of various target objects; then, based on the full analysis of the image of the target object space, texture, spectral information characteristics, selected a large amount of information, related information of characteristic small again; the target objects, the knowledge base is established according to the actual situation in the study area, to create semantic knowledge rules, to achieve accurate extraction of Eucalyptus information; finally, to realize the eucalyptus forest remote sensing information extraction accuracy evaluation.The results and conclusions of this paper are as follows:(1) Image analysis method based on object oriented application in high resolution images than the traditional pixel based image analysis method has more advantages. The consideration of high resolution images in spectral, texture, attribute space characteristics, can use the feature information of feature rich image multi scale segmentation, build target objects, and establish corresponding knowledge base and knowledge reasoning rules, realize the accurate classification of image objects. The study found that: the object oriented classification method is more suitable for high resolution remote sensing image information extraction, more targeted, more targeted, more accurate information extraction results.(2) Remote sensing information sources on different levels, different space, different spectral resolution is analyzed, various characteristics of TM, Spot, Ikonos/QuickBird, contrast resource satellite, RapidEye, the spatial resolution of RapidEye data, the digital processing capability, price ratio advantage of Eucalyptus forest monitoring data meet the requirements, and the red edge the band can monitor the vegetation growth.(3) The core problem of object-oriented image analysis method is accurate image segmentation, extraction of feature information in image recognition and correlation characteristics are the basis of image segmentation, image segmentation quality will directly affect the subsequent object information feature extraction, target recognition and identification results. The study found that: image segmentation compared to the larger targets, implement larger scale segmentation, and the use of smaller scale segmentation, forest segmentation scale with 60 ~ 75, the tree object segmentation using 40 ~ 50; spectral factor setting control algorithm parameters is 0.7, compactness is set to 0.5, the better the effect of image segmentation.(4) Facing the key technical analysis method of the object image, in many of the selected object features a large amount of information, information characteristics of small factor correlation, and find out the correct implementation of object feature difference threshold classification. The study found that, using the feature selection method based on coefficient of variation and correlation coefficient, according to the standard deviation and the simple correlation coefficient of object features in the preliminary selection, the final selection of expansion factor, get the heterogeneity, homogeneity, angle two moments of 11 informative than larger feature types, for object classification effect.(5) To evaluate the accuracy of information extraction of eucalyptus trees results, the calculation method of confusion matrix, GPS combined with field survey data, to test the classification results. The results showed that: the classification results of object oriented technology and the overall accuracy reaches 82.12%, Kappa coefficient was 0.8; while the traditional supervised classification results based on pixels of the overall accuracy is 68.83%, Kappa coefficient is 0.67, extraction of eucalyptus trees and classification of object oriented technology is more suitable for Eucalyptus forest information, higher precision.(6) Image analysis method based on object oriented technology, the remote sensing information extraction application in Eucalyptus forest has the advantages of high efficiency, objective, accurate, effective, tracking and monitoring can realize the dynamic information of Eucalyptus planting, growth, distribution etc.. In the Eucalyptus classification process, can use inheritance mechanism, fuzzy logic concepts and methods, and the semantic structure, create a eucalyptus forest information extraction rules, can eliminate the small difference spectra of small or mixed pixels caused by speckle. The extraction can also be applied to other surface information of object oriented technology.