Ammatzia Peled
University of Haifa, Israel
BACKGROUND and MOTIVATION
Credible classification of remotely sensed data is one of the most studied issues in the geo-information science. This challenging issue incorporates various scientific disciplines (such as geodesy, computer sciences, geography etc.) in order to collect and process digital data regarding various terrain phenomena. Although these phenomena vary in size and extent, it is difficult to separate them one from the other. The significant advance in remotely sensed data quality and its availability, enable the study of a large number of phenomena, simultaneously. In addition, there is a rapid growth of available data sources (more satellites, new airborne scanners, advanced ground scanners and field surveys). This multitude of available data has assisted to overcome traditionally problematic issues such as incomplete data, absence of data and contradictions between observations. On the other hand, the increase of these multi-source complex data necessitates the automatization of image understanding processes.
TOPICS
Introduction
Updating needs and strategies
Unsupervised and Supervised classification
CAL/VAL
Standard Conditions Concept and meteorological Models
Training Sub-sets
Training sets
Sampling
Control and Check sets
Peled GIS-Driven Concept
Information embedded in Spatial data bases
No Cal/VAL
No standard conditions
Any Sensor
GIS-Driven Updating of GIS data Bases
Thematic Line-following
Quality Control
Rule-Based
Knowledge-Bases
Summary
Achievements and Drawbaks
What is missing
Autonomous Updating