Introduction of Infomax Learning Algorithm and Application for Oil Spill Detection in SAR Images

抄録

It is useful method to detect targets by extracting difference in local patterns of Remote Sensing images. However, it has not been established yet. Especially, in case that the pattern has difficulty in representing mathematical description, automatic detection is difficult as well. In this case, the pattern extraction by experienced people visually is the most effective method at present. Therefore, we consider neural network as a model of human visual recognition process, and propose automatic pattern extraction method featured by neural network learning based on Infomax Learning Algorithm. We applied this method in extracting pattern of oil spills on the sea observed by SAR images, as an example, and demonstrate that target can be detected automatically. In order to save computer load in learning and detecting, process, we approximate the feature of these patterns by Gabor function and demonstrate that the target can be detected as almost same level. Hence, we demonstrate that the proposed method is effective for automatic target detection.