Geo-Interpretation Model for Land-Cover/Land-Use Classification
Symbolic Knowledge Reasoning Based Land-Cover/Land-Use Classification
Recently, in the research and application of remote sensing land-cover/land-use classification, the knowledge based logic reasoning classification method or even to build expert system have been carried out and have got fruit progressing. A symbolic knowledge based system may be a large and complex set pf program functioning as an expert system that simulates the higher-order interpretative processes of human analysts When such knowledge representation imitates of simulates the logic of human perception, it is called machine or computer vision.
With the support of symbolic geo-knowledge base, including spectral knowledge base, topological knowledge base and land-use knowledge based, and logic reasoning machine, the classification outcome of land-cover can be validated and modified by obeying to the facts and rules in the realistic land cover and land use distribution. By building the spectral knowledge base, S.W.Wharton(1987)improved the accuracy in urban land-cover classification. B.Kartikeyan (1995) built the frame work of remote sensing land-cover classification expert system including spectral knowledge base, reasoning machine, knowledge automatic acquisition unit, by which the example work land-cover classification cal carried out.
Based on conventional geo-analysis thought of geo-difference in geo-phenomenon, symbolic knowledge based logic reasoning remote sensing classification method realizes the discriminance in remote sensing image by symbolic representing to the geo-knowledge and formalization reasoning machine. Thus, in some degree it can truly reflect the geo-distribution in the remote sensing image. But for the characteristics of RS information being fuzzy and complicated, it is difficult to represent the land-cover regularities and duynami9cal growing process hidden in remote sensing by structural and symbolic geo-knowledge. Further more, for remote sending image generally comprise huge information, the efficiency should nesting site low when using serial symbolic reasoning forms to processing and analyzing the image data.
Remote Sensing Intelligent Geo-Interpretation Modal
In this section, we will propose a classification system named RSIGIM/ Remote Sensing Intelligent Geo-Interpretation Modal (RSIGIM) is a hierarchical processing and analyzing modal to research how to simulate the synthetically Geo-expert's interpretation and decision process to the remote sensing image by computer system. From law level to high level, RSIGIM passes through several courses including information transfers, statistical basic processing and analyzing unit, physiological vision cognition, psychological logic cognition, knowledge discovery, decision and analysis. The process is just like the intelligent interpretation process of an expert to picture from receiving, observation, judgment, association to memory . (Figure1).
Figure 1: The Framework of RS Intelligent Geo-Interpretation Model
There are several targets of to present out the RSIGEM as below.
-
To describe, recognize, classify and interpret the geo-object, phenomenon and manner hidden in remote sensing image;
- To extract the imagery mechanism and intrinsic feature including classifying type, size, structure, relationship and others geo-properties of geo-objects in RS image;
- To discover and represent the geo-knowledge from huge accumulated image base;
- And further more, to do prediction and decision to the geo-phenomenon and geo-manner by being integrated by certain geo-analysis model.
Based on the RSIGIM, we present out the intelligent land-cover/land-use classification system, which includes three levels' structure models and methods (fig2). The first or low level is the statistical RS image processing and analyzing model, which is based on the parametric or non- parametric distribution model to the original processing and analyzing working by mathematical computing and statistical analyzing methods, such as MLC, clustering, etc. IN This level, by the statically difference in feature space pf unknown or known object distributed in RS image the rough information classification or extraction is realized to attain the primary target of cognition to the distribution of basic units in the RS image. The middle level is neural computing model based classification model, which is to simulate human's visual thinking of the vision neural processing system in order to realize the hierarchical land cover classification by firstly learning the samples datum with supported by defini8te geo-knowledge to get the complex non-linear mapping network structure. The typical neural computing models ads algorithms on pattern classification include BP-MLP,RBF, ARTMAP,etc. The third level, also being the highest level, is the symbolic logic reasoning mode. In this level, we firstly should build the structural geo-knowledge base, such sc spectral knowledge bane, topological knowledge base, in which the knowledge is represented by symbolic forms, typically like RULE-"IF -THEN. Then by logic reasoning machine the output of land cover classification is validated, updated and modified obeying the facts or geo-regularities represented or hidden in knowledge base ,and more further the land use map is updated under the condition of land use and more further the land -use map is update under the condition of land-use map acquired form manual vision interpretation by expert.
Figure 2 Systematic Structure of land -cover classification based on RSIGIM
