|
|
|
| The clustering of high resolution remote sensing imagery |
| Deng Xiangjin; Wang Yanping; Peng Hailiang |
| Institute of Electronics Chinese Academy of Sciences Beijing 100080 China |
|
|
|
|
Abstract The technology of clustering high resolution imagery is difficult, due to the fact that the minor components, such as roads, make the appearance of the same category region non-uniform. This paper proposes a new approach to cluster high resolution remote sensing imagery. The clustering approach includes three steps. First, eliminate the minor components in moving windows. The process uses 1-D morphological watershed technique to find the left threshold and the right threshold in the histogram. The gray levels beyond the two thresholds which result from minor components will replaced by the principle gray level. This process can improve the statistic measures when the moving windows contain some small hetero-objects. Second, compute the image characteristics in moving windows. Third, apply BPC neural network, which is combined by a back-propagation network and a competitive network, to cluster images according to the images characteristics. Three approaches are tested using SPOT images for clustering residential areas and agricultural areas in the suburb of Beijing. The experimental results show that the new clustering approach has the highest clustering accuracy.
|
|
Received: 24 January 2002
|
|
|
|
|
|
|
|
|
2003, Vol. 25 
