Object-based Extraction Of Moso Bamboo Forest And Estimation Of Moso Bamboo Inventory Factors

Bamboo is an important forest resource in China, which has abundant ecological, economic andesthetic value. Accurate mapping of bamboo information is a fundamental step to monitor bambooresource and estimate the carbon storage using remotely sensed data. Processing units in object-basedapproach are objects rather than pixels in the conventional pixel-based classification approach.Object-based classification approach is applied not only relying on the spectral information ofremotely sensed data, but also the spatial information, such as geographical location, spatial extent,texture, and so on. This technique can reduce the “salt and pepper” effect caused by the variation ofthe spectral responses in the same entity, especially for the very high-spatial resolution (VHR)imagery in which the same entity is usually represented by pixels with high spectral heterogeneity.The study area is Shanchuan town, which is located in the south of Anji county. The objectives ofthis thesis were to: delineate Moso bamboo (Phyllostachys heterocycla var. pubescens) forest usingobject-based method, which provided the advantages of multi-scale segmentation and developinghierarchical structure, combined with textural information; integrate multi-source images into theobject-based classification and select the most appropriate scale for image segmentation; and toestimate Moso bamboo inventory parameters based on the combination of object-oriented anddecision tree mothod. It mainly contains the following aspects:1. Multi-scale, object-based extraction of Moso bamboo forest. Based on SPOT5remotelysensed imagery, Moso bamboo forest was delineated using two object-based schemes, which providedthe advantages of multi-scale segmentation and developing hierarchical structure. Finally eitherclassification results of the two object-based schemes and the conventional pixel-based maximumlikelihood method was compared.2. Synergistic use of Landsat TM and SPOT5imagery for object-based Moso bambooclassification. Synergistic use of Landsat5TM and SPOT5images was evaluated for improving forestclassification using an object-based image analysis approach, including design of multi-scalesegmentation schemes, selection of the optimal segmentation scales and the accuracy comparison ofMoso bamboo forest and other forests.3. Estimation of Moso bamboo inventory parameters based object-oriented and decision tree algorithm. The SPOT5image was segmented into image objects using the multi-scale segmentationmethod, and the features of objects, such as spectral, texture and layer characteristics, were extracted.Then, the rules were automatically obtained using classification and regression tree (CART) algorithm.Classification tree was carried out to develop the rules for Moso bamboo extraction. At last,regression tree was used to identify correlation between feature of image objects and Moso bambooinventory parameters of field-derived samples.Through this present study, conclusions can be drawn as follows:1. The most appropriate window sizes for calculating texture using red (R), green (G) and blue(B) band in SPOT5image were9×9，7×7，9×9; Extraction of Moso bamboo using object-basedmethod yielded the result with more accuracy, with the producer’s accuracy reaching90%, obviouslyhigher than that of the conventional maximum likelihood method (88.57%); Multiresolutionsegmentation with the aid of texture not only ensured the high accuracy of Moso bamboo, but alsoprovided help to the other forest types. The overall accuracy was92%and the Kappa coefficient was0.8814, both of which were the highest accuracy in the present study.2. The optimal scales for the segmentation of TM/SPOT5, SPOT5and TM were70,100and0.8,respectively; classification results with medium spatial resolution images were not desirable, withoverall accuracy of only72.35%, while synergistic use of Landsat5TM and SPOT5greatly improvedforest classification accuracy, with overall accuracy achieving82.94%.3. The overall accuracy of diameter at breast height (DBH) and height model results basedsub-compartment were40.74%,66.67%, with Kappa of0.0182,0.0471, which proved to beunreliable. The overall accuracy of diameter at breast height (DBH) and height model results basedfield-derived samples were both50%, with Kappa of0.3222,0.2308. Thus, the model results ofsamples was better than that of sub-compartment on the whole, which may give insight into theretrieval of Moso bamboo inventory parameters using optical remote sensing data.