Algorithm Study On Object Oriented Classification For Coastal Zone Landform

With large amount of information, access to information faster, shorter cycle, less restricted and so on, remote sensing provides a new method for the investigation of the coastal zone. Image interpretation of coastal zone is the most common application on coastal zone investigation. Currently, object-oriented classification has become new direction of remote sensing image interpretation. That method firstly segment image to object, which is a series of homogeneous regions adjacent to each other. After that, classify these objects. Compared with traditional method based on pixel, object contain more shape and texture information which is able to describe image feature more comprehensively so as to improve the classification accuracy.In recent years, some scholars have applied the object-oriented classification to the classification of coastal landform. However, coastal landform is complex where natural features and artificial features exist simultaneously. Thus spectrum, texture, shape and distribution characteristics of costal landform is complicate. Current algorithms are unable to extract all kind of features. Which means there is a lot of space to be improved upon object oriented classification of coastal zone. Hence, following work is done in this paper:a) Determine feature class according to the characteristic of coastal zone image. Propose an object oriented classification method for coastal landform based on texton by the application of texton theory.b) The structure tensor is used to optimize the object-oriented classification based on texton theory:Trace of structure tensor is used to obtain the gradient image to optimize image segmentation; structure tensor texture is used to construct the texton to optimize the classification.c) Local Binary Patterns (LBP) is a simple and effective texture algorithm. The author focus on the application of LBP on extraction of coastal landform. Generate marker image in view of LBP and doing watershed algorithm. Finally, mining geometric information of the object and classify the objects using BP neural network.As a consequence, when employing image spectral information, texton also taps image texture information and can better reflect the characteristics of image. Thus, Object Oriented Classification Based on Texton Theory can be effectively applied in classifying coastal landforms. Moreover, Structure can optimize the process of segmentation and clustering, so as to improve the whole classification process. Besides, there is quite big room for application of LBP on coastal landform classification for it have a high classification accuracy on extraction of some coastal landforms.