Research On Water System Auto Classification

The paper conclude current classification methods based on visual interpretation;discuss the water system's classification gist by visual interpretation;analyze the possibility and advantage classified by computer;put forward the classification method using the whole water system texture as a classification cell which is different from traditional method using a window as a classification cell.On the basis of Remote Sensing Intelligent Geo-interpreting Model(RSIGIM), simulate the process of classifying water system styles by human's brain, then symbolize the experts' thinking and abstract the cognizing factors. Discard traditional method that use a fixed window to be the research cell, but make the water system's extending range to be the research cell. Study the computer classification method by combining texture encoding and pattern recognition technology. Make a whole water system as a basic cell, discuss the encoding method, classification factor and classification criterion used in water system classification. The technical course of the water system classification prototype system is that: first, to get the data source, transact the image by binarization algorithm and thinning algorithm based on mathematical morphology;encode the water system texture with 8-direction code because the water system image is grid image;then, simulate human brain's thinking, fix 6 distinguishing factors to classify the water systems, they are branch direction, branch angle, branch length, whether the branch normal pass the water system's center area, whether the direction from water system center to branch end is the same with the direction of the branch, and how many first class branches join in one second class branch;calculate these 6 factors with the 8-direction code we have;last, according to the classification method and the 6 factors, judge each water system's style. When classifying, if we find one water system contains more than one style, we will divide it into several sub-water systems by the division module.The prototype system is made of 5 modules: water system obtaining, water system texture encoding, classification factor calculation, water system classification and water system division. The system can discriminate 7 different kinds of water systems, they are dendritic water system, parallel water system, grid-like water system, orbicular water system, radialized water system, fanlike water system and penniform water system.Input 34 water system images interpreted from RS images and classify these images by the system;the average ratio of correctness is beyond 80%.last, the paper lists the validity of each type.