Evaluation of a hybrid symbiotic system on segmenting SAR imagery

For many problems, a knowledge-based approach towards analyzing remotely sensed data, like SAR imagery, is appropriate. System organization is top-down, with low-level processing routines at the bottom and high-level symbolization and knowledge processing at the top. Such systems assume that a given problem can be solved in the context of a system´s low-level processing routines. In other words, such systems assemble solutions in layers of “chunks”, where chunks in lower layers consist of image processing algorithms, while chunks in higher layers consist of artificial intelligence methods. However, not all image analysis problems can be approached in a strictly top-down fashion, especially if significant uncertainties exist in understanding how a geophysical or biological event maps to a radar backscatter signature. A subsequent approach would relax a top-down hierarchy and would allow for mixing of symbolic, knowledge, and image processing chunks. Consequently, the first author has been developing a method for mixing symbolization and low-level processing chunks, especially when the processing chunks become smaller than an algorithm but larger than a word of code. J.M. Daida (1994) describes specific implementations of this method for SAR image segmentation. This paper briefly describes this method and evaluates the performance of one of the implementations for segmenting ERS-1 and JERS-1 imagery