Study On Discrimination Of Oil Slicks And Lookalikes In Polarimetric SAR Images Using CNN

Oil slicks and lookalikes(e.g.,plant oil and oil emulsion)all appear as dark areas in polarimetric Synthetic Aperture Radar(polSAR)images and are highly heterogeneous,so it is very difficult to use a single feature to classify oil slicks and lookalikes.In this paper,combination of multi feature fusion and convolution neural network(CNN)is used to support the discrimination of oil slicks and lookalikes in full polarimetric SAR images.The main work of this article is summarized as follows:Firstly,the 18 characteristics of the oil slicks and lookalikes are extracted and analyzed,and the features which have good distinguishability finally rationalized a preferred features subset.According to the results of the experimental analysis,the characteristics of the preferred features subset include entropy,alpha,and Single-bounce Eigenvalue Relative Difference(SERD)in the C-band polarimetric mode.Then,the regions of interest(ROI)are selected as the training and testing samples for CNN on the three kinds of polarimetric feature images.The proposed method is applied to a training data set of 5400 samples,including 1800 crude oil,1800 plant oil,and 1800 oil emulsion samples.The classification accuracy obtained using 900 samples(300 crude oil,300 plant oil,and 300 oil emulsion samples)of test data is 91.33%.Finally,the artificial neural network(ANN)classification algorithm is used to classify the oil slick and lookalike area based on the same experimental data,and compared with the classification results of CNN classification algorithm.Considering that the over fitting of the network model should be avoided in order to obtain more reliable experimental results,the K-cross validation and the ROC curve experiment are also carried out in this paper.Through experiments,we can find that under the condition of insufficient experimental data,there will be an over fitting phenomenon in the experimental models.Therefore,we should pay attention to the enrichment and expansion of experimental data in subsequent research.The effectiveness of the method is demonstrated through the analysis of some experimental results.It is here observed that the proposed method not only can accurately identify the dark spots on SAR images but also verify the ability of the proposed algorithm to classify unstructured features.