Change detection in SAR imagery

Two separate SAR image change detection schemes are described. In the absence of ground truth one must judge these schemes against the performance that may be achieved by eye. On this basis, both schemes are found to work well. The neutral network approach appears to perform slightly better than the model-based scheme in terms of both time and detection rates. Given the fact that the author has the option of training the networks on a more realistic object set than that currently used it would appear that the network approach is to be preferred. It is estimated that both schemes will run in real-time on a new transputer-based vector processor