Data Fusion, the Core Technology for Future On-board Data Processing System

Currently, more and more earth observation data have been acquired by many kinds of sensors on different platform, such as optic sensors, microwave sensors, infrared sensors, hyperspectral sensors, etc. Thanks to giant resource being required to store and transmit these tremendous data so that the cost is very large and the efficiency is low, investigators are compelled to process them on-board as possible as they can. So far, on-board data processing only settles on some simple preprocessing, such as correction, denoising, compensation, etc. Information extraction not only is the objective of earth observation, but can distill large amount data so that amount of data needing to be stored and transmitted is reduced greatly. Feature extraction, change detection, and object recognition executed on-board will provide us an efficient information extraction system for earth observation. Data fusion technique has been widely used to process earth observation data on the ground, which can generate data with higher quality and extract better information from multisource or multitemporal data. Furthermore, data fusion can also be used to extract better information from these data on-board, simultaneously, the redundant data will be eliminated greatly so as to accelerate data processing and reduce data for storage and transmission. However, on-board data fusion processing will confront more difficulty, one of the most principal troubles is that on-board data processing system must be completely autonomous, which results in some procedures such as image registration, feature extraction, change detection, object recognition becoming more complicated, while they can be processed by help of manual operates despite being difficult on the ground. Of course, the tremendous advantage of data fusion for on-board data processing will promote investigators to remove the obstacles on the road to on-board data fusion-based information extraction. 1. CURRENT STATUS, TREND AND STRATEGIC DIRECTION OF ON-BOARD DATA PROCESSING With the rapid development of information technology, the users’ requirement to information is transform-ing from static state, un-real-time to dynamic state and real-time. The dynamic state and real-time information acquired by remote sensing technology has been used successfully in the area of urban planning, precision farming, vegetation coverage, ocean observation, disaster monitoring, etc. How to utilize the great amount data and the higher and higher resolution for the earth observation has been a focus to the users and investigators. Currently, earth observation data are always processed after being transmitted to ground data processing center (GDPC), then, in GDPC, the data are de-noised, and corrected geometric and radiometric bias, and then accurate images can be acquired. After that, image classification, object category, target recognition, and decision generation are executed. Finally the processed result will be distributed to the entity units. For the sake of earth’s curvature, sometimes the earth observation data and processed result are transmitted on the help of the ground delay station. From foresaid procedure we can conclude that data processing and maintaining are very complicated because of the complex linking route, data type, and data format in the traditional earth observation data pro-cessing mode. Moreover, distribution of the processed data is also an intricate problem, sometimes the processed results need to be transmitted to the communicating satellite, and then distributed to the end users. Not only does the increase of number and depth of the linking route result in the increase of the complexity of data maintain-ing, but it reduces the reliability of the data processing system, and also increases the processing and transmit-ting time, going against real-time data processing and distribution. Furthermore, with rapid increase of the earth observation data quantity and image resolution, users are more and more eager to acquire information to detect objects’ change, recognize target.