Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Remote sensing technology > Interpretation, identification and processing of remote sensing images > Image processing methods

Configured Targets Recognition in High-Resolution Remote Sensing Image

Author DuBo
Tutor TanYiHua
School Huazhong University of Science and Technology
Course Pattern Recognition and Intelligent Systems
Keywords target-recognition configured high-resolution remote sensing images image segmentation
CLC TP751
Type Master's thesis
Year 2010
Downloads 119
Quotes 0
Download Dissertation

With resolution of remote sensing images becoming increasingly higher and information becoming richer, there is the possibility to take roads, buildings and airports as automatically recognition targets. Through the research on widely applied automatic target recognition technology, we find it plays an important role in civil navigation and military targets with precision strikes etc. In recent years, many researchers have done a great deal of work in this area. However, the technology of automatically extracting information from the remote sensing images and recognizing the target is still not perfect and mature. Nowadays, most automatic target recognition technology can only be used to recognize certain target without universal value.Taking example for airports, ships, buildings, roads and other typical remote-sensing targets, the paper tries to research on configured object segmentation and recognition and introduce object-oriented segmentation and recognition method. The main idea is: The image is divided into object units by segmentation. Segmentation and recognition algorithms are made independent. Based on appropriate rules, object units are configured with segmentation method to obtain targets. Then, by adopting feature-configured recognition, automatic recognition of various targets can be achieved. Features of this paper are as follows:Firstly, in the object segmentation, the paper studies object-oriented segmentation in which the method can be configured. The main idea is: the image is divided into a series of object units, from which targets to be recognized can be obtained by post-processing. According to target characteristics, the paper introduces three candidate ways to do object segmentation preprocessing. Taking airports, ships, buildings and roads as examples, it also presents the configuration results of object segmentation methods.Secondly, in the feature extraction, the paper describes in detail feature extraction algorithm based on high-resolution remote sensing target. In-depth mining several major characteristics of the object, such as spectral, texture, geometry, and these sufficient candidate characteristics can be provided for object recognition with configured features. At last, in the target identification, the paper focuses on the target recognition technology with configured features. It puts forward a feature configuration rule based on separability measure. AdaBoost classifier is used to study features that have been configured to identify the classification threshold, the weight coefficient and the configuration to reach identifiable parameters for each feature. In the end, taking ship as an example, a comparasion between target recognition algorithms and support vector machine recognition algorithms is made. The study shows that target recognition rate rises through feature configuration.

Related Dissertations
More Dissertations