A new maximum likelihood classification technique for multitemporal SAR and multiband optical images

In this paper we devise a new fusion technique for a sequence of multitemporal single-channel SAR images of a given area with a single multiband optical image. Unlike for SAR, the availability of optical images is largely affected by atmospheric conditions, so that this is a case of practical interest. First, a statistical model for the joint distribution of SAR and optical data is provided. Then the corresponding joint maximum likelihood (ML) classifier is derived, and lower and upper bounds to classification performance are introduced. An optimized technique for ML joint image segmentation and classification is proposed, showing results close to the upper bound. Finally the effectiveness of fusion of SAR and optical images is investigated quantitatively, showing a consistent performance improvement with respect to using either sensor alone.