A scheme of intelligent oil spill monitoring system by SAR for operational application is presented. The system includes four key techniques. 1) Picking out dark targets from SAR images using a new adaptive threshold segmentation algorithm based on evaluating the trends of background of SAR image along the range direction. The algorithm is applicable for SAR images of different satellites. 2) Filtering out some of look-alikes from dark targets by a chain of rules. 3) Extracting features from all the remained dark targets, selecting the most useful features and then discriminating the targets between oil spill and look-alikes by an artificial neural network (ANN). The feature extraction is based on lots of targets. The ANN experiences enough training. 4) Using an intelligent feedback with expert knowledge and relevant environment parameters to continually optimize system and improve the detection rate. So far, the test on 1448 oil spill and look-alike targets from Envisat/ASAR images shows that the correct recognition rate of the system can reach 88 % without the intelligent feedback.
展开▼
