The importance of the extraction and cognition of geo-information has been increasingly highlighted in the face of the massive accumulation of remote sensing data and the lack of application information. According to the geo-informatics Tupu methodology regarding the visual cognitive process

Tupu-cognition can automatically interpret remotely sensed imagery. In this study

using a unified framework of geographic information systems

we extract the features of images step by step. Spatial-spectrum a nalysis is then executed in the geo-cognitive process described as "Perceive-Identify-Confirm ". Algorithms like multiscale s egmentation

feature analysis

and supervised learning are invoked to meet the application’s requirements for automation and intelligence. In the cognitive application of land-cover information

we first establish the mechanism of prior knowledge management for automation. Second

a number of machine learning algorithms are employed to improve the intelligence. Finally

adaptive iteration is introduced to optimize the results. The data selected for this classification experiment are Advanced Land Observing Satellite（ ALOS） multispectral images in the Pearl River Delta. The land cover results are consistent with expectations and illustrate the f easibility of our method.