Conclusion

Summary

We have described the system for the structural analysis of complex aerial photographs. This system is an image-understanding system which automatically locates a variety of objects using diverse knowledge of the world. Several experimental results have shown that it works fairly well for various types of scenes of urban and suburban areas.

In Chapter 1, we pointed out several shortcomings in the ordinary statistical classification and target detection, and stressed the necessity of introducing artificial intelligence techniques into the analysis of remote-sensing imagery. Our system is one of the first that attempts to automate the process of photointerpretation of complex aerial photographs.

Automatic photointerpretation by computer includes many different aspects of the problem: picture processing, feature extraction, knowledge representation, and control structure. Several sophisticated picture processing techniques have been newly developed for structuring the raw picture data. They have helped the system to transform a two-dimensional array of observed data into a wellorganized data structure. Many experiments were performed for extracting features useful for the discrimination of objects. These features have facilitated the reliable recognition of diverse objects in aerial photographs. The primary focus of this research has been on how to organize individual picture processing techniques and diverse knowledge sources in a flexible way. Several new ideas have been introduced for realizing an efficient and reliable analysis: the focusing mechanism for estimating approximate areas of objects, the production system architecture for representing the diverse knowledge, the feedback (heterarchical) analysis for locating context-sensitive objects, and the feedback correction of segmentation errors.